Impact of Adjusted and Nonadjusted Surface Observations on the Cold Season Performance of the Canadian Precipitation Analysis (CaPA) System

Pei-Ning Feng aCentre ESCER, Department of Earth and Atmospheric Sciences, Université du Québec à Montréal, Montreal, Quebec, Canada

Search for other papers by Pei-Ning Feng in
Current site
Google Scholar
PubMed
Close
https://orcid.org/0000-0003-4647-2372
,
Stéphane Bélair bAtmospheric Science and Technology, Environment and Climate Change Canada, Dorval, Quebec, Canada

Search for other papers by Stéphane Bélair in
Current site
Google Scholar
PubMed
Close
,
Dikraa Khedhaouiria bAtmospheric Science and Technology, Environment and Climate Change Canada, Dorval, Quebec, Canada

Search for other papers by Dikraa Khedhaouiria in
Current site
Google Scholar
PubMed
Close
,
Franck Lespinas cCanadian Centre for Meteorological and Environmental Prediction, Environment and Climate Change Canada, Dorval, Quebec, Canada

Search for other papers by Franck Lespinas in
Current site
Google Scholar
PubMed
Close
,
Eva Mekis dClimate Research Division, Environment and Climate Change Canada, Downsview, Ontario, Canada

Search for other papers by Eva Mekis in
Current site
Google Scholar
PubMed
Close
, and
Julie M. Thériault aCentre ESCER, Department of Earth and Atmospheric Sciences, Université du Québec à Montréal, Montreal, Quebec, Canada

Search for other papers by Julie M. Thériault in
Current site
Google Scholar
PubMed
Close
Open access

Abstract

The Canadian Precipitation Analysis System (CaPA) is an operational system that uses a combination of weather gauge and ground-based radar measurements together with short-term forecasts from a numerical weather model to provide near-real-time estimates of 6- and 24-h precipitation amounts. During the winter season, many gauge measurements are rejected by the CaPA quality control process because of the wind-induced undercatch for solid precipitation. The goal of this study is to improve the precipitation estimates over central Canada during the winter seasons from 2019 to 2022. Two approaches were tested. First, the quality control procedure in CaPA has been relaxed to increase the number of surface observations assimilated. Second, the automatic solid precipitation measurements were adjusted using a universal transfer function to compensate for the undercatch problem. Although increasing the wind speed threshold resulted in lower amounts and worse biases in frequency, the overall precipitation estimates are improved as the equitable threat score is improved because of a substantial decrease in the false alarm ratio, which compensates the degradation of the probability of detection. The increase of solid precipitation amounts using a transfer function improves the biases in both frequency and amounts and the probability of detection for all precipitation thresholds. However, the false alarm ratio deteriorates for large thresholds. The statistics vary from year to year, but an overall improvement is demonstrated by increasing the number of stations and adjusting the solid precipitation amounts for wind speed undercatch.

© 2023 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Pei-Ning Feng, feng.pei-ning@uqam.ca

Abstract

The Canadian Precipitation Analysis System (CaPA) is an operational system that uses a combination of weather gauge and ground-based radar measurements together with short-term forecasts from a numerical weather model to provide near-real-time estimates of 6- and 24-h precipitation amounts. During the winter season, many gauge measurements are rejected by the CaPA quality control process because of the wind-induced undercatch for solid precipitation. The goal of this study is to improve the precipitation estimates over central Canada during the winter seasons from 2019 to 2022. Two approaches were tested. First, the quality control procedure in CaPA has been relaxed to increase the number of surface observations assimilated. Second, the automatic solid precipitation measurements were adjusted using a universal transfer function to compensate for the undercatch problem. Although increasing the wind speed threshold resulted in lower amounts and worse biases in frequency, the overall precipitation estimates are improved as the equitable threat score is improved because of a substantial decrease in the false alarm ratio, which compensates the degradation of the probability of detection. The increase of solid precipitation amounts using a transfer function improves the biases in both frequency and amounts and the probability of detection for all precipitation thresholds. However, the false alarm ratio deteriorates for large thresholds. The statistics vary from year to year, but an overall improvement is demonstrated by increasing the number of stations and adjusting the solid precipitation amounts for wind speed undercatch.

© 2023 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Pei-Ning Feng, feng.pei-ning@uqam.ca

1. Introduction

Accurate precipitation estimation is challenging. This is especially true in Canada, a vast country bound by three oceans, spanning from the midlatitudes to the polar region. One of the main reasons is the sparse spatial distribution of observation stations over the country (Mekis et al. 2018). Other difficulties relate to the varying topography, cold climatic conditions, and the general lack of easy road access, making it difficult to maintain weather stations. The improvement of precipitation products across Canada is then crucial.

To account for this gap in precipitation information across Canada, the Canadian Precipitation Analysis (CaPA) system has been developed at Environment and Climate Change Canada (ECCC) since 2003 (see Mahfouf et al. 2007). CaPA combines a background field built from short-range forecasts provided by a numerical weather prediction (NWP) system and observations from weather stations and radars to produce near-real-time gridded estimates of precipitation amounts for 6- and 24-h accumulation periods (Mahfouf et al. 2007; Fortin et al. 2018; Mekis et al. 2018). Lespinas et al. (2015) provide a comprehensive overview of CaPA’s extensive evaluation, whereas Fortin et al. (2018) summarize its evolution through the years.

Many observational datasets are used in CaPA, from surface-based to space-based observations. At the surface, this includes provincial and national networks, as well as private sector networks. Mekis et al. (2018) provides an overview of ECCC’s national surface precipitation observations. To measure precipitation, many types of gauges are used for surface stations, but some are more reliable in cold environmental conditions (Rasmussen et al. 2012). For instance, weighting gauges are mainly used to measure snowfall, but their collection efficiency decreases with increasing wind speed. Such findings were obtained by comparing measurements from several gauge-shield types with those from reference gauge-shield configuration, which is an automatic gauge placed in the double fence intercomparison reference (DFIR; Thériault et al. 2015; Kochendorfer et al. 2022).

The wind-induced undercatch is caused by the air deflection in the vicinity of the gauge orifice, leading to an updraft upstream of it. This weak upward motion prevents some of the precipitation to enter the gauge (e.g., Rasmussen et al. 2012; Thériault et al. 2012a,b; Colli et al. 2015). When compared with rain, the trajectories of solid precipitation are more affected by the near-surface wind and the ensuing updraft (Nešpor and Sevruk 1999; Thériault et al. 2012a,b). For instance, slow-falling particles are more likely to miss the gauge than the fast-falling particles (e.g., Leroux et al. 2021; Hoover et al. 2021).

To account for the precipitation gauge undercatch, the WMO Solid Precipitation Intercomparison Experiment (SPICE) was organized (Goodison et al. 1998). Various gauge-shield intercomparisons were conducted to develop transfer functions (Kochendorfer et al. 2022). Transfer functions were developed to adjust solid precipitation amount as a function of wind speed and air temperature measured by an automatic gauge placed in a single Alter shield. This type of adjustment from transfer functions has been shown to improve snowfall estimates (Kochendorfer et al. 2017a,b).

Given the difficulties in measuring solid precipitation, current operational versions of CaPA assimilate far fewer ground observations during the cold season than during the warm season. For air temperature below the freezing point, a large part of the observations from surface stations are rejected in CaPA because of its quality control. Furthermore, radar-derived precipitation is only assimilated in CaPA for rainfall (see Fortin et al. 2015); precipitation estimates in colder situations, more likely to fall as solid precipitation, are not used at this moment (see also Fortin et al. 2018). The goal of this study is thus to demonstrate that precipitation analyses from CaPA can be improved by relaxing its wintertime quality control, i.e., by increasing the number of assimilated observations from surface stations, and by adjusting the amount of solid precipitation using a universal transfer function.

The paper is organized as follows: section 2 provides a description of the surface observations networks assimilated in CaPA, the analysis system, and the experimental design of the experiments; section 3 presents the results of the two experiments carried out in this study, in comparison with the operational control; and section 4 provides an analysis and discussion of critical aspects of this study, and the summary, along with a perspective on future work, is presented in section 5.

2. Data and model description

a. The Canadian Precipitation Analysis (CaPA) system

The CaPA system was developed by ECCC to provide real-time precipitation analyses for both 6- and 24-h periods (Mahfouf et al. 2007; Fortin et al. 2018). The precipitation analyses are obtained by combining a first guess built from short-range NWP with several types of observations, including surface measurements, ground-based radars, and space-based satellite products (Boluwade et al. 2017; Fortin et al. 2018). The combination of observations and NWP short-range outputs is performed using an optimal interpolation technique in which error statistics are derived from a variographic analysis (Evans 2013; Fortin et al. 2018).

In this study, a version of CaPA that only assimilates observations from surface meteorological stations is applied (Lespinas et al. 2015; Mekis et al. 2018). Ground-based radars and satellite products are not utilized. Also, this work is based on a 2.5-km version of CaPA, available since 2016 (Fortin et al. 2018), which builds on a background field from ECCC’s High-Resolution Deterministic Precipitation System (HRDPS; Milbrandt et al. 2016), a national 2.5-km configuration of the Global Environment Multiscale (GEM) model (e.g., Côté et al. 1998; Milbrandt et al. 2016; McTaggart-Cowan et al. 2019).

The domain of interest covers a region in Canada that includes Alberta, Saskatchewan, Manitoba, Ontario and parts of British Columbia, the Northwest Territories, and parts of the north-central United States (Fig. 1). The domain contains several surface stations that provide precipitation data to CaPA, as described in section 2b. Note that this study’s analysis domain is different from CaPA’s national 2.5-km system run operationally at ECCC (covering the entire country). Other than the practical reasons related to the computational cost of running such a system at the national scale for several years and experiments, the choice of this area is motivated by the relatively uniform topography and orography found over the Canadian prairies. This geographical area also responds to a special interest for using CaPA’s wintertime precipitation analyses for hydrological modeling over central Canada. Although the studied domain includes parts of the Rocky Mountains in British Columbia, there are very few manual stations used for objective evaluation over that area, thus minimizing the impact that mountain areas could have on the results and analyses presented below.

Fig. 1.
Fig. 1.

Analysis domain of the study (black-outlined frame) together with the location of surface stations considered for assimilation in CaPA (dots), for a specific date (0600 UTC 7 Jan 2019). The colors are the various surface station networks used in this study. For the topography, the colors are the surface covered by plants (green) and the exposed land surface (beige).

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

The experiments in this study were done for 6-h analyses only. The main reason is that transfer functions can now be used in CaPA to adjust the precipitation accumulations, but only based on the average wind speed over the entire accumulation period. This makes their use for 24-h accumulations questionable since transfer functions were built for shorter (i.e., 30-min) events. Thus, each day, four 6-h analyses are produced at synoptic times 0000, 0600, 1200, and 1800 UTC. For each analysis, the cumulative precipitation of the forecast between the 6- and 12-h lead times generated by the HRDPS 12 h before the validity date of the analysis is used as the first guess (Khedhaouiria et al. 2020).

b. Surface precipitation observations

As indicated in Fig. 1, the surface observations assimilated into CaPA come from several networks of meteorological stations, which are equipped with different types of gauges. These networks include aviation stations, automatic stations (from ECCC and from other sources), network partner stations, manual stations, and superstations. Each type of station is identified by a given code type number, allowing them to be treated separately as part of CaPA’s quality control procedures. Super stations, with a specific code type, correspond to artificial stations that are created by merging stations located close to each other (Lespinas et al. 2015). ECCC’s automatic stations are from Surface Weather and Reference Climate Stations (RCS; Mekis et al. 2018), with reporting time not limited to synoptic hours. Figure 1 shows the location of surface observations (stations) assimilated by CaPA for a specific date, i.e., 0600 UTC 7 January 2019. The number of observations at stations assimilated in CaPA varies for each validity date, depending on data availability and on the quality control applied on that date.

Over the Canadian portion of the analysis domain, CaPA assimilates observations from ECCC’s weather stations as well as stations belonging to provincial partners including British Columbia Wildfire Service (BC Forest) and the Ontario Ministry of Natural Resources and Forestry (OMNRF). The networks of partner stations are located over various regions of the analysis domain. They include automatic weather stations and manual climate stations. They are operated by the partner organizations and provide data to ECCC.

In the United States, CaPA assimilates observations from the synoptic network (SYNOP), which are daily data and therefore not used in this study of the 6-h accumulation, but most of the observations are in fact from the American Meteorological Terminal Aviation Routine (METAR), which are observational data collected from stations located at airports. This network includes about 1800 stations with availability from hourly to daily. The weighing gauges are the most common in this network and report precipitation events only when the precipitation exceeds zero (Lespinas et al. 2015; Mekis et al. 2018).

The utilization of the equipment for the observation of solid precipitation varies from network to network (Mekis et al. 2018). For the periods examined in this study, the surface observation networks changed in a few ways. Some manual stations became automatic, and more network partner stations were added to the surface network. As Table 1 shows, the average of the daily number of total observations increased from 2019 to 2022. Winter is defined in this study as December–March. As an example, the period from December 2018 to March 2019 is defined here as winter 2019.

Table 1.

Average daily number of observations of each winter (from 2019 to 2022) that are assimilated and rejected by the CaPA system. CTRL is the control run, QC is the quality control experiment, and TF is the experiment of the application of the transfer function.

Table 1.

c. Adjustment of solid precipitation

Because of difficulties in accurately measuring solid precipitation in windy conditions with precipitation gauges, the impact of adjusting surface observations to compensate for gauge undercatch is investigated. The universal transfer function developed within the Intercomparison WMO Solid Precipitation Experiment (WMO SPICE; 2012–15; Nitu et al. 2018) is tested in this study. The transfer function development methodology is described in Kochendorfer et al. (2017a), with the catch efficiency defined as
CE=ea(U)(1{tan1[b(Tair)+c]}).
The transfer function considers the effect of wind speed U (m s−1) and of air temperature Tair (K). The coefficients a, b, and c applied here are respectively 0.0348, 1.366, and 0.779 (Kochendorfer et al. 2017a, their Table 2), fitting to the single Alter shielded weighing gauge surface data using 1.5-m wind observations. This universal transfer function is designed to correct the undercatch bias from the measurement of all the similarly shielded precipitation gauges (Kochendorfer et al. 2017a, 2018). To avoid additional uncertainty associated with mixed precipitation events, the transfer function adjustment for solid precipitation is defined for Tair < −2°C as suggested in Kochendorfer et al. (2017a).

Although the universal transfer function of Kochendorfer et al. (2017a) was designed for the correction of the 30-min precipitation measurements, the reference also mentions the equal impact it has on observations with 60-min time interval. Furthermore, the application of the transfer function to larger time intervals as long as 12 and 24 h is also widely illustrated as feasible (Wolff et al. 2015). In the present study, the universal transfer function is implemented in CaPA for 6-h precipitation analyses.

d. Experimental design

The CaPA assimilation cycles are initialized on 1 November and end on 31 March for a series of four winter seasons, from 2019 to 2022. For each winter, the first month (November) serves as a spinup period to stabilize the error statistics used in CaPA’s optimal interpolation, which are updated for each 6-h assimilation cycle. The period from 1 December to 31 March is used to assess and compare the quality of analyses between experiments. No distinction is made in this study about the precipitation phase, i.e., the total precipitation from CaPA is objectively evaluated, and not rain and snow separately.

The configuration of the control run (CTRL) in this study closely resembles the operational High-Resolution Deterministic Precipitation Analysis (HRDPA; see Khedhaouiria et al. 2020; CMC 2018) system employed by ECCC. One of the main differences with HRDPA is the smaller computational domain [see Fig. 1 and cf. with Fig. 6 of CMC (2021), i.e., all of Canada]. Also, ground-based radar observations are not included here. In Canada, the quality control wind speed threshold for meteorological situations with air temperature below −2°C is 3.0 m s−1 for manual stations, and 0.6 m s−1 for the automatic stations. Observations for situations with wind speed greater than the threshold are rejected from the assimilation process. In the United States, the wind speed threshold is 2.0 m s−1 for the manual stations and 0.6 m s−1 for the automatic stations (this discrepancy between the United States and Canada is inherited from the operational CaPA system).

The first experiment, referred to as “QC,” allows for more surface observations to pass through the quality control process and be assimilated in CaPA. This is achieved by increasing the wind speed threshold from 0.6 to 3.0 m s−1 for the automatic stations in Canada; all other parameters remain the same as for CTRL. The wind speed threshold of 3.0 m s−1 for near-surface conditions aligns with findings from prior research.

The study of Nitu et al. (2018) highlights a significant decline in catch efficiency when wind speeds exceed this threshold. In their comprehensive experiments, numerous gauges spanning various stations and gauge types were examined for their catch efficiency. While the results exhibited some variability across different stations and gauge configurations, it consistently revealed that the wind speed threshold of 3.0 m s−1 for near-surface conditions consistently improves catch efficiency. While variations exist across stations and gauge types, this threshold generally outperforms others. In specific cases, higher thresholds may offer similar efficiency, but they may not align with actual precipitation events. Therefore, the conclusion of Nitu et al. (2018) emphasizes that a wind speed threshold of 3 m s−1 is optimal for measurements, especially when analyzing frequency relative to wind speeds both above and below this threshold.

The default setting of wind speed threshold for the assimilation of Canadian manual stations in CaPA is 3.0 m s−1 as well. In the present study, Fig. 2 shows that about 80% of observations from automatic stations are assimilated with this wind speed threshold. In comparison, this percentage is less than 20% in CTRL with the wind speed threshold as 0.6 m s−1. Therefore, over 60% of total observations from automatic stations are added into CaPA in QC.

Fig. 2.
Fig. 2.

Number of assimilated observations from automatic stations as a function of 1.5-m wind speed threshold for each winter season (2019–22).

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

The impact of the transfer function is assessed in the second experiment, named “TF.” This experiment has the same configuration as for QC but the universal transfer function is implemented to adjust measurements from automatic stations to correct for undercatch. Importantly, TF and QC share the same number of observations, thus facilitating interpretation of the results. In this study, the adjustment in TF is applied to precipitation observations when Tair < −2°C to only account for solid precipitation; the same wind speed threshold as QC is used for TF’s quality control. The adjusted precipitation is obtained by dividing the measured precipitation amount by the catch efficiency [Eq. (1)].

e. Objective evaluation

Objective evaluation of precipitation analyses is challenging because a large portion of available observations is used in the assimilation process. Evaluation is even more difficult in the cold season because of the gauge undercatch problem described above. The most used method to evaluate CaPA’s precipitation analyses is based on the leave-one-out (LOO) approach (e.g., Friesen et al. 2017; Fortin et al. 2015; Lespinas et al. 2015).

Following this method, used in this study, CaPA’s optimal interpolation analysis is performed at the locations of surface stations using neighboring observations, i.e., without considering the observation taken at this location. The LOO analysis values are then compared with the observations at these locations (which were excluded in the LOO analysis process), and objective evaluation is then produced based on the categorical metrics described below. In CaPA, only manual synoptic observations that pass the CaPA quality control are used for the LOO evaluation for the winter season.

Several verification metrics are calculated to assess the quality of the precipitation analyses generated by CaPA. These are the frequency bias index (FBI), the equitable threat score (ETS), the probability of detection (POD), and the false alarm ratio (FAR). The process to classify events as hits, false alarm, and miss is illustrated in Table 2. A success is a situation for which CaPA correctly predicts the observation, a miss is when CaPA fails to capture an observed event, and a false alarm is when CaPA predicts an event that did not occur. The total number of observed events is the sum of hits and misses, whereas the total number of predicted events is the sum of hits and false alarms.

Table 2.

Definition of the events for the verification metrics.

Table 2.

FBI is the measurement of the ratio of the number of predicted events to the number of observed events:
FBI=hits+falsealarmshits+misses.
In this study, FBI minus 1 (hereinafter FBI-1) is shown and evaluated. Values of FBI-1 > 0 indicate that CaPA’s analyses tend to overestimate the number of events for a specific threshold, and the opposite when FBI-1 < 0. The optimal value of FBI-1 is 0.
The ETS measures the fraction of analysis values that are correctly predicted, adjusted by the number of hits expected by chance due to climatology. ETS ranges from −⅓ to 1, with values of ETS = 1 representing a perfect prediction and values of ETS < 0 for cases with no skill. The ETS is computed as follows:
ETS=hitsrandomsuccesshits+misses+falsealarmsrandomsuccess
in which the random success is
randomsuccess=(hits+misses)(hits+falsealarms)N
with N as the total number of events. More detailed explanations for both FBI and ETS can be found in Lespinas et al. (2015) and Ferro and Stephenson (2011).
The POD is a measure of the ratio of the observational events that are successfully predicted. It is computed as follows:
POD=hitshits+misses.
The ideal value for POD is 1. As compared with ETS, POD ignores the false alarms and is not adjusted for random success. This verification metric is thus more sensitive to climatological frequency and is better interpreted when examined together with FAR, which is defined as
FAR=falsealarmshits+falsealarms.
The FAR is the ratio of the predicted events that have been predicted but have not been observed. The optimal value for FAR is 0, which means that there are no incorrect predictions. Like POD, FAR is also sensitive to climatological frequency.
In addition to these metrics scores, partial means of precipitation (PM) are calculated for all experiments and compared with observations. This metric provides a measure of how well the statistical distributions of CaPA’s precipitation analyses fit with those of surface observations in terms of intensity. PM is an effective tool to assess how the precipitation mass from the analysis compares to that observed at the gauges for specific intensities. The PMs for CaPA’s analyses xA¯(τ) and for surface observations xo¯(τ) are evaluated as follows:
xA¯(τ)= xA|xA<τ={ xA|xA(m)<τ }xANo.{ xA(m)<τ }and
xo¯(τ)=xo|xo<τ={ xo|xo(m)< τ }xoNo.{ xo(m)<τ },
in which τ is the 6-hourly precipitation threshold (mm), xo(m) is the observed value at spatiotemporal location m, and xA(m) is the analysis from the LOO experiment at the same location. Note that as τ increases and reaches large values the PM tends toward the “total” precipitation amount and essentially provides information on the precipitation bias (all intensities included).

In this study, the stationary block bootstrapping is applied to compute the confidence interval during the verification process and to estimate the statistical significance of the metric differences between the experiments (section 3a in Lespinas et al. 2015).

3. Results

a. Case study

During the four winters considered in this study, a few precipitation events were selected as case studies to illustrate the impact of adding more surface observations in CaPA and of adjusting automatic observations with a transfer function. After removing from consideration cases for which precipitation occurs over areas not covered by surface observations, located in the Rocky Mountains for instance or in the very northern areas of the study domain, the precipitation analysis valid at 0600 UTC 7 January 2019 was selected as an example and is shown in Fig. 3.

Fig. 3.
Fig. 3.

Domain of the study and the precipitation analysis at 0600 UTC 7 Jan 2019 produced by the (a) first-guess, (c) CTRL, (e) QC, and (g) TF experiments with the spatial distribution of the assimilated (orange) and rejected (blue) observations. CaPA’s analysis domain for this study is represented by the shaded area. (b) The difference in precipitation amount between CTRL and the first guess. (d) The difference in precipitation amount and the location of observations (stations) assimilated in CTRL and QC. (f) The average value of the wind speed (uv) measured at the stations over the 6-h accumulation period. The difference in precipitation amount between CTRL and QC are the shaded areas in (d) and (e). (h) The difference in precipitation amount between the QC and TF analyses. For (b), (d), (f), and (h), the contour interval is 0.5 mm.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

To illustrate the impact of the two modifications investigated in this study, with the QC and TF experiments, and to display the spatial distribution of the surface observations assimilated in CaPA, Fig. 3 shows the 6-hourly precipitation from the model first guess (Figs. 3a,b), the influence of increasing the number of surface-based assimilated observations (between CTRL and QC, Figs. 3c–f), and the impact of adjusting surface observations with a transfer function (between QC and TF, Figs. 3g,h).

In all the analyses and estimates for that case, precipitation occurs primarily in the southern part of Manitoba, and the northern and middle areas of Saskatchewan. Weaker precipitation is also found over northern Alberta and partially in British Columbia. When compared with the first guess shown in Fig. 3a, precipitation patterns appear as being smoothed by the analysis process (Fig. 3b). The largest impact of the assimilation of surface observations is for a smaller-scale event at the border of Alberta and Saskatchewan. The dipole of positive and negative increments (red and blue patches) suggest that the event was moved northward because of surface observations. The same trend over that location is found with the inclusion of more surface observations, as shown by the QC and CTRL maps (Figs. 3c–e). In Figs. 3c and 3d for CTRL and QC, the number of surface observations increases from 101 to 261 when the wind speed threshold for automatic station is increased from 0.6 to 3.0 m s−1. The number of rejected observations decreases correspondingly. The newly assimilated observations are distributed over the entire studied domain: the increase in the number of assimilated observations is greatest in southern Alberta, where station density is very high, and lowest in Saskatchewan and Manitoba, where station density is lower.

As the number of surface observations increases, CaPA’s precipitation analysis for that date is also modified. Notably, the CaPA analysis for QC tends to reduce precipitation accumulation in most areas, especially those where precipitation is larger than 6 mm (Figs. 3a,c). Where the most significant discrepancies occur corresponds to the precipitation areas with the newly integrated assimilated observations in QC, as expected (Fig. 3d). The amplitude of the differences between QC and CTRL is larger in general than that of the differences between CTRL and first guess.

For that specific case, the precipitation analysis for areas with higher wind speed generally has larger values in CTRL, which is mainly determined by the first guess because no observations are assimilated at these locations. Assimilating more observations in the QC experiment results in a significant reduction of precipitation amounts in the analysis in many regions (Figs. 3b,d). For other areas the contribution is reversed; that is, adding surface observations has the impact of increasing the precipitation analysis (Fig. 3b). The amplitude of this correction is, however, smaller. No determination is made at this point on the quality and accuracy of these precipitation estimates. Figure 3 only illustrates the effect of the two modifications investigated in this study.

The wind speed spatial distribution with categories based on the thresholds used for the CTRL and QC experiments is shown Fig. 3d. The spatial distribution of wind speed categories is generally consistent with that of the assimilated observations for all experiments (Figs. 3a–c). The wind speed threshold of 0.6 m s−1 in CTRL only allows for the assimilation of surface observations in a few areas (Fig. 3a), whereas the threshold of 3 m s−1 in QC increases the number of assimilated observations by a factor of 2.5.

The precipitation analysis of the TF experiment has spatial patterns similar to those of CTRL and QC (Fig. 3e). To evaluate the impact of the transfer function, the differences between the QC and TF precipitation analyses are shown in Fig. 3f. It shows that the transfer function adjustment systematically increases (by design because the CE is always lower than 1) the precipitation amounts, with the largest increase in southeastern Manitoba by a maximum amplitude of about 2 mm.

The overall influence of the transfer function (Fig. 3f) for this specific weather event is weaker than the impact of increasing the number of surface observations (Fig. 3b). The QC experiment leads to greater corrections in heavy precipitation areas. In contrast, TF is less correlated to the precipitation locations and more with the wind speed. Thus, QC leads to greater corrections, up to 3 mm when compared with CTRL, whereas the impact of TF with respect to QC is at most 2 mm.

b. Impact of relaxed wind speed criteria in CaPA quality control (QC experiment)

When using many cases, the verification metrics help to evaluate CaPA’s precipitation analyses objectively. Here, the evaluation was performed individually and combined for the four different winters in this study.

The QC has a similar impact with respect to CTRL for each of the four winters. The inclusion of more surface observations has the effect of decreasing the FBI-1 value (Fig. 4). For all the winters, CaPA underestimates the number of events with accumulation greater than 1 mm. The amplitude of the FBI-1 difference between CTRL and QC is also in a similar range for all the winters, that is, about 0.15–0.2.

Fig. 4.
Fig. 4.

The frequency bias index FBI-1 valid for the study domain for CTRL (blue) and QC (red) (a)–(d) for each winter and (e) for the four-winter summary. Here, q is the 6-hourly precipitation accumulation (mm). Filled symbols mean that the differences between CTRL and QC are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

During the 2019 winter (Fig. 4a), both CTRL and QC have the FBI-1 negative for all the precipitation thresholds. According to the definition in section 2, both CTRL and QC underestimate the frequency of precipitation events for small and large amounts. The differences in FBI-1 are statistically significant at the 95% confidence level for all the accumulation thresholds.

For the other three winter seasons, from 2020 to 2022, the impact of QC is similar, with a substantial decrease in the FBI-1 values when compared with CTRL for all the thresholds. This impact for the three seasons is an improvement for smaller precipitation amounts (thresholds of 0.2, 0.5, and sometimes 1.0 mm h−1).

The results are summarized in the four-winter average presented in Fig. 4e, which shows in a consistent manner the very systematic impact on precipitation bias. For small precipitation thresholds, CaPA tends to overestimate precipitation frequency, and the opposite for thresholds greater than 1 mm. For precipitation accumulations greater than 2 mm, the FBI-1 decreases for both CTRL and QC. Although FBI-1 was overall underestimated in winter 2019, the average as presented in Fig. 4e remains similar to the general pattern of the other three winters. When all the precipitation events are taken into the metrics calculations, the sample size gets larger than a single winter. Hence, the statistical comparison is also strengthened. The FBI-1 thus indicates that when the upper limit of the wind speed for automatic stations is raised from 0.6 to 3.0 m s−1 as the condition for quality control, the impact on the CaPA analyses is mainly to significantly decrease the precipitation frequency for these four winters.

The partial means shown in Fig. 5 provide a different view of the modifications to systematic differences (biases) related to the assimilation of more surface observations. Consistent with FBI-1, the partial means indicate that precipitation amounts in the QC analyses are lower than those from CTRL for all the winters, in a statistically significant manner. The impact is positive or negative, depending on the year under evaluation. For winter 2019 (Fig. 5a), the impact of QC versus CTRL is negative, since both experiments produce precipitation analyses that underestimate the observational precipitation amounts, as evidenced by the asymptotic values reached for high precipitation accumulations. This is consistent with the negative FBI-1 in Fig. 4a. The situation is different for the other three winters, from 2020 to 2022, for which precipitation analyses from CTRL have higher partial means than the observations, while QC analyses are closer to the observations.

Fig. 5.
Fig. 5.

As in Fig. 4, but for partial mean.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

The 4-yr average reflects this dual impact (Fig. 5e), with QC partial means slightly smaller than observations for the precipitation accumulations larger 5 mm, with a positive impact for accumulations less than or equal to 5 mm, and with a more neutral impact for greater accumulations. When considering total precipitation, CTRL overestimates precipitation amounts while QC underestimates them by about the same quantity in both directions.

The ETS differences shown in Fig. 6 between the two experiments are less systematic than for FBI-1 and partial mean. For winter 2019 (Fig. 6a), the ETS for CTRL and QC are close to each other and do not differ in a statistically significant way, which indicates that the accuracy of precipitation analyses does not change substantially when more observations are assimilated in CaPA. There are some differences for the following winters 2020–22 (Figs. 6b–d), with QC exhibiting better ETS scores than CTRL. The ETS for QC is significantly higher than CTRL for thresholds below 1.0 mm. For higher thresholds, the sample size decreases and the differences of ETS between CTRL and QC are not statistically significant. The four-winter average (Fig. 6e) shows that ETS for small precipitation events (<2.0 mm) is significantly higher for QC than for CTRL. As the accumulation threshold increases, the ETS for both CTRL and QC drops to values lower than 0.40 and the difference becomes less apparent and are statistically not significant.

Fig. 6.
Fig. 6.

As in Fig. 4, but for equitable threat score (ETS).

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

The ETS depends both on the system’s ability to detect events and on the false alarms. For the detection aspect, the POD values show that when more observations with higher wind speed limits are assimilated in QC, the POD becomes lower than CTRL (Fig. 7). This is consistent with decreased values of biases, previously discussed with FBI-1 and partial means (Figs. 3 and 4). This decrease in POD is found consistently for all the winters considered in this study. The four-year average clearly shows a deterioration of detection, with statistically significant differences for all but the largest precipitation threshold, that is, 5 mm (Fig. 7e).

Fig. 7.
Fig. 7.

As in Fig. 4, but for the probability of detection (POD).

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

Similarly, in a mirror effect and in a manner consistent with the bias performance, Fig. 7 shows a clear reduction in the number of false alarms in QC when compared with CTRL. This improvement is found for each winter season and is statistically significant for most events for all the winter seasons. The FAR value for winter 2019 is about 0.3 for most precipitation thresholds (Fig. 8a), which is lower than for other winters, with FAR values about 0.4 (Figs. 8b–d). The FAR values are typically consistent for small threshold values, then decrease slightly for the 2.0-mm threshold, and then increase for the largest threshold used here, 5 mm. The consistent pattern from each winter therefore leads to similar FAR values when considering the four winters together (Fig. 8e) and shows a convincing difference between the CTRL and QC experiments.

Fig. 8.
Fig. 8.

As in Fig. 4, but for the false alarm rate (FAR).

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

c. Impact of adjusting with a universal transfer function (TF experiment)

As a follow-up to the QC experiment described in the previous section, adjustments were made to the surface observations from automatic gauges based on the universal transfer function described in section 2c. The evaluation metrics of CaPA for the comparison between QC and TF are presented in this section to illustrate the effect of the universal transfer function. In CaPA’s LOO approach, the evaluation metrics are calculated using observations from stations that are shared by two experiments, for the comparison to be as fair as possible.

The impact of the precipitation adjustment from the transfer function is shown Fig. 9. The precipitation values assimilated in CaPA as part of TF are always equal or larger than what is used in QC. Consequently, the FBI-1 scores are higher for TF than for QC for all the winters (Fig. 9). The differences in the FBI-1 scores between the two experiments, however, are less than 0.1. The FBI-1 values remain negative even with the increase in precipitation amounts using the TF, except for the small precipitation events in the winters of 2021 and 2022, indicating a precipitation underestimation for both QC and TF.

Fig. 9.
Fig. 9.

The frequency bias index FBI-1 valid for the study domain for QC (red) and TF (purple) (a)–(d) for each winter and (e) for the four-winter summary. Here, q is the 6-hourly precipitation accumulation (mm). Filled symbols mean that the differences between QC and TF are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

The FBI-1 scores calculated by considering all cases from the four winters together are shown in Fig. 9e. Because of the negative FBI-1 scores for the first two winters, the FBI-1 scores for the four-winter summary remain negative for all the precipitation thresholds. For small precipitation events up to the accumulation threshold of 1.0 mm, FBI-1 for both QC and TF is close to 0. The FBI-1 decreases as the accumulation threshold increases, indicating again that CaPA more severely underestimates larger precipitation events. Applying the transfer function adjustment to increase precipitation amounts is statistically significant for most precipitation thresholds, even though this increase is insufficient to adequately address the negative biases of CaPA for large precipitation events (Fig. 9e).

The precipitation partial means for the selected winters and the four-winter summary are shown in Fig. 10. Consistent with FBI-1, the partial means for TF are higher than those of QC for all winters and for all precipitation thresholds. This is also the case for the four-winter partial means shown in Fig. 10e. The overall negative FBI-1 values of winter 2019 (Fig. 9a) are consistent with lower partial means for both TF and QC when compared with observations (Fig. 9a). For all the precipitation thresholds, the increases in partial means with TF are statistically significant. The other three winters show a similar signal, with overestimation of the partial means for intermediate thresholds (e.g., 2.0 mm) and asymptotic values that are relatively close to the observations.

Fig. 10.
Fig. 10.

As in Fig. 9, but for precipitation partial means.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

Figure 10e shows that the partial means calculated by considering all cases of the four winters together increases significantly for events greater than 5.0 mm with TF, and the total precipitation mass approaches that of the observations. The four-winter partial means for TF are influenced by the underestimation found in winter 2019. Therefore, they are slightly lower than the observations for thresholds larger than 5.0 mm but higher than QC as for each winter.

Overall, the ETS values for TF are slightly better than QC for all winters and most precipitation thresholds, as shown in Fig. 11. There is some year-to-year variability for the ETS and the impact of the adjusting precipitation observations using the transfer function. For instance, the ETS values are generally lower for winter 2020 than for the other winters, with values around 0.40 and smaller. The ETS values for TF are systematically higher than for QC for all winters, but these differences are not statistically significant. The ETS values computed from the four winters show that the signal is more stable in terms of the TF and QC differences but also in terms of statistical significance (Fig. 11e). Except for the precipitation threshold of 0.5 mm, the improvement resulting from the adjustment is clear and statistically significant.

Fig. 11.
Fig. 11.

As in Fig. 9, but for ETS.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

POD scores are statistically higher for TF than for QC for all precipitation thresholds and for all winters, although the magnitude of the increase is less than 0.05. Similarly, the POD scores for the period of all four winters combined show the same substantial improvement (Fig. 12). The POD pattern for each winter shown in Fig. 12 has some similarities to the ETS (Fig. 11). It is found that the application of the transfer function to the POD yields a greater effect relative to the ETS after discarding false alarms and random successes.

Fig. 12.
Fig. 12.

As in Fig. 9, but for POD.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

An increase in bias and POD is often accompanied by a deterioration (i.e., increase) in FAR, which is observed when comparing this metric for TF and QC (Fig. 13). In general, FAR is relatively constant across precipitation thresholds (on the order of 0.4), except for winter 2019, which has a lower FAR with values between 0.20 and 0.30. This is consistent with the lower values of FBI-1 and partial means for that year (Figs. 8a and 9a). The FAR values remain relatively unchanged with TF for thresholds lower than or equal to 0.5 mm while they are increased for more intense precipitation events. These differences are observed for each winter as well as for the period of the four winters combined and are statistically significant for the 1.0- and 2.0-mm thresholds.

Fig. 13.
Fig. 13.

As in Fig. 9, but for FAR.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

The combined impact of adding surface observations and adjusting observations from automatic surface stations (all included in TF), relative to the original control experiment (CTRL), is shown in Fig. 14. All four winters combined for all evaluation metrics are analyzed.

Fig. 14.
Fig. 14.

Evaluation of CTRL (blue) and TF (purple) for the period of the four winters combined. The y axis is the evaluation metric, and the x axis is the 6-hourly precipitation accumulation threshold (mm). Filled symbols mean that the differences between CTRL and TF are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

One of the main features regarding the impact of both changes is related to the bias. Even if the universal TF has the systematic effect of increasing precipitation (through its catch efficiency), the overall impact of TF relative to CTRL is to reduce precipitation, both in amount and frequency. This reduction is due to the increase in the amount of precipitation observations for surface stations, which lowers biases as previously shown in the comparison between QC and CTRL. This is evidenced by the substantial decrease in FBI-1 and partial mean values (Figs. 14a,b). The FBI-1 values for TF are close to zero and slightly negative for precipitation thresholds lower or equal to 1 mm and are lower for larger thresholds [5 mm (6 h)−1 and more]. The partial means also reveal that precipitation accumulations are reduced with TF, indicating an improvement over CTRL, especially for larger thresholds close to asymptotic values. The total precipitation values obtained with TF are slightly lower than the station observations, while CTRL significantly overestimates the total precipitation amount.

The combination of both changes also has a substantial impact on the ETS (Fig. 14c). The differences in ETS that were obtained between QC and TF (Figs. 5 and 10) are now evident when comparing TF and CTRL, with large and statistically significant improvement for small thresholds (smaller or equal to 1 mm), similar to what was found for QC. The impact is also positive for larger thresholds (2 and 5 mm), as was found with the QC experiment, although the differences between TF and CTRL are not statistically significant. The POD and FAR scores reveal that the gain in ETS is primarily associated with decreased FAR for the TF experiment. POD is very similar, except for a slight deterioration for the smallest thresholds (Figs. 14d,e).

4. Discussion

We examine in this study potential ways to enhance the impact of assimilating observations from surface stations on wintertime precipitation analyses produced by CaPA over central Canada. The first test is performed by relaxing the quality control currently used in all of CaPA’s operational configuration to include surface observations from automatic stations. This is achieved by increasing the upper limit of the acceptable wind speed in cold conditions, with air temperature below −2°C. In that experiment (named QC), the threshold value in CaPA’s configuration settings is modified for automatic stations from 0.6 m s−1 in the control experiment (called CTRL) to 3.0 m s−1 for experiment QC. The higher value for this wind speed parameter allows for more observations to be assimilated into CaPA. The second experiment, named TF, has the same setup as experiment QC but a universal transfer function is applied to adjust (increase) solid precipitation observed at the surface from automatic stations.

Increasing the number of surface observations assimilated in CaPA during several winter seasons (experiment QC), from 2019 to 2022, has the overall effect of decreasing precipitation in CaPA’s analyses, with a mixed impact on biases both in terms of events frequency (FBI-1) and in terms of precipitation amount (partial means). For FBI-1, the systematic decrease has a deteriorating impact on larger precipitation accumulations; the effect is more mixed for weaker precipitation events. The results are also mixed for the partial mean and total precipitation, but still slightly in favor of QC with departures from observations that are reduced for several precipitation thresholds, and with total precipitation (indicated by the asymptotic values) of equal quality. In CTRL, the partial mean exhibits a distinct wet bias that seems to surpass the dry bias caused by the undercatch of the gauge observations, indicating that the assimilation of observations from surface gauges continues to be beneficial for the studied winters. Values of both detection (POD) and false alarms (FAR) are consequently decreased, leading to an overall positive impact on the ETS for smaller precipitation thresholds.

Although not as pronounced when compared with the QC experiment, the increase of precipitation amounts associated with the TF experiment also positively impacts CaPA’s precipitation wintertime analyses over central Canada. The bias (FBI-1 and partial means) is slightly improved, with a slight increase in detection and false alarms. Even though it is small, the overall effect on ETS is positive and statistically significant for most precipitation thresholds.

The combined effect of the two modifications, obtained by comparing the TF and CTRL experiments, reveals an overall positive impact. For the bias, the decrease associated with QC is more dominant than the increase associated with the adjustment from TF, leading to a general reduction for both FBI-1 and partial means when comparing TF with CTRL. This can be considered an improvement for the partial means. Still, the conclusions are more mixed for the FBI-1, i.e., with only a slight improvement for small thresholds and an apparent deterioration for larger thresholds. In terms of accuracy, gains for both QC and TF are evident when comparing ETS with the CTRL experiment. All thresholds are improved, but the impact is larger for weaker events. The benefits from the modifications presented here are mostly linked with a decrease in false alarms. At the same time, detection remains primarily unchanged, but it is only slightly worse for smaller thresholds.

The ability of CaPA to estimate 6-hourly precipitation accumulations varies from year to year. This is also true for the impact that the two modifications tested in this study has on CaPA’s performance. The first winter season examined in this work, i.e., winter 2019, appears to stand out when compared with the last three years of the study, i.e., from 2020 to 2022. The main differences, as shown in Figs. 413, are related to precipitation biases as shown with the FBI-1 scores and partial means. Specifically, both FBI-1 and partial means are much lower in 2019 than in other years.

For the first winter, the operational configuration of CTRL underestimates precipitation in terms of both its frequency (FBI-1) and amounts (partial means). Deeper examination of the results indicate that this might be at least partially explained by the quality of the first guess. Figure 15 shows FBI-1 for CTRL and the first guess for each winter. For winter 2019 (Fig. 15a), the first guess tends to underestimate the precipitation for most of the thresholds, in contrast with the other three winter seasons for which FBI-1 is above the zero line for the 0.2-, 0.5-, and 1.0-mm thresholds. For these other years, the assimilation of surface observations lowers FBI-1 and thus corrects this overestimation (for these smaller thresholds), whereas the effect is different for 2019 for which the frequency bias is worsened (lowered).

Fig. 15.
Fig. 15.

The frequency bias index FBI-1 valid for the study domain for CTRL (blue) and the first guess (green) (a)–(d) for each winter and (e) for the four-winter summary. Here, q is the 6-hourly precipitation accumulation (mm). Filled symbols mean that the differences between CTRL and QC are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

We mention that the partial means and the FAR (not shown) also indicates that the first guess is better for winter 2019 relative to the other three years. For the partial mean, the first guess of winter 2019 approaches the observations the best. For the other years, the first guess tends to overestimate the partial means, a feature that is partially corrected with the assimilation of surface observations. In contrast, the first-guess FAR for winter 2019 is in general lower than 0.4, better than for the other years in which the first-guess FAR values are all higher than 0.4. Again, the inclusion of additional surface observations reduces this problem associated with the first guess.

It is likely that this temporal variability of one of CaPA’s main contributors (first guess) largely determines the nature of the impact of the changes tested in both the QC and TF experiments. It seems that for winter 2019 the negative effect on FBI-1 and partial sums balances other positive impacts associated with adding surface observations, which could be related for instance with intensity and position of precipitation events. The consequence is that the performance of both QC and TF in terms of ETS is more similar to CTRL for that specific winter season.

Other reasons could be responsible for this interannual variability. It may be related, for instance, to the nature and level of weather activity over central Canada in each of these winter seasons. To inform on this, the total precipitation over each winter season (from 1 December to 31 March) from the model background field is shown in Fig. 16. Other than the Rocky Mountains area (Fig. 1), the spatial distribution and amount of precipitation vary substantially from winter to winter. The 2019 winter season is drier than the other years over the northern region of the analysis domain. Over the southern parts, other years have more extreme seasonal totals. Winter 2022 is found to be the wettest, with precipitation totals over 100 mm for a large portion of the domain. Winter 2021 is the driest, with large areas with totals less than 50 mm. Winter 2019, on the other hand, is similar to 2020 over these areas with more intermediate precipitation seasonal totals. This year-to-year variability does not seem to further explain why the impact of adding surface observations is less for winter 2019.

Fig. 16.
Fig. 16.

The total precipitation of the background field (the first guess) from HRDPS of each winter.

Citation: Journal of Hydrometeorology 25, 1; 10.1175/JHM-D-23-0070.1

Another reason for this variability could be related to the composition of the surface observing network, i.e., the number of surface-based observations from observers (manual) and from automatic stations might vary from year to year as some of the manual observations were converted during those years into automatic stations. Figure 1 already shows the sparse spatial distribution of the manual stations used for the objective evaluation. In that context, an evaluation over portions of the analysis domain might not be statistically significant due to the relatively small number of manual stations. Table 1 already revealed that the total number of surface observations did not change much for these four years. It is possible, however, that some of the manual stations could have been converted into automatic stations during the study period, which could have an influence on the objective evaluation. Again, it seems unlikely that these small modifications to the surface observing network is the main cause of the different impact seen in 2019, especially considering the argument presented below related to the model first guess.

Another aspect of this study requiring further understanding and explanations is related to the approach used for CaPA’s objective evaluation, now based on a LOO method with assimilated manual observations using an upper-limit wind speed limit of 3.0 m s−1 when the air temperature is below −2°C. Based on this fact, events with stronger wind speeds, often associated with more intense precipitation (solid or liquid), could be underrepresented in CaPA’s evaluation process. This potential bias toward weaker events (as done at ECCC for CaPA’s objective evaluation for its operational implementation) could impact the metrics presented here. Objective evaluation versus other datasets, such as the Adjusted Daily Rainfall Snowfall dataset (e.g., Wang et al. 2017), is being developed for future studies investigating CaPA’s winter season performance.

Other studies are currently being conducted to further improve the wintertime performance of CaPA. They include possible refinements to the precipitation adjustment using more appropriate transfer functions for this specific area and methods to increase the observation uncertainty associated with the transfer function adjustment (not done here). Additional experiments are under way to examine the impact of other transfer functions, including during the winter season retrievals from ground-based radars and the Integrated Multi-Satellite Retrievals for Global Precipitation Measurement (IMERG; Asong et al. 2017).

5. Summary

In this study, three experiments of the period from winter 2019 to 2022 are conducted to evaluate CaPA’s precipitation analyses for 1) the control run configuration as CTRL, which is close to ECCC’s HRDPA system, 2) the QC experiment in which the wind speed threshold for Canada’s automatic stations is modified from 0.6 m s−1 (as in CTRL to 3.0 m s−1), and 3) the TF experiment with the imposition of the universal transfer function to automatic stations under −2°C.

In QC, more observations from automatic surface stations are assimilated into CaPA’s analyses. Based on a LOO objective evaluation using observations from manual surface stations, results show that QC produces precipitation analyses that mitigate the overestimation problem noticed in CTRL. The partial mean is closer to observations with QC than CTRL in most cases. The POD in the QC experiment is decreased, but both FBI-1 and the FAR are lowered (and generally improved) in a manner more substantial than POD’s decrease, with an overall positive impact on ETS. This is an important result that shows that even with the undercatch problem associated with snowfall measurements by automatic gauges, the assimilation of unadjusted observations still has a positive impact on CaPA’s analyses.

As expected, the impact of TF is to increase the precipitation amount because of its correction associated with undercatch of surface automatic gauges. With this adjustment in TF, the FBI-1, the partial mean, and the POD are all increased accordingly. The FAR is also higher in the 4-yr summary. Even though it is still positive, the amplitude of TF impact relative to QC for the LOO objective evaluation is smaller than the impact of just adding the observations, i.e., by comparing QC with CTRL. The ETS increase for the two modifications to CaPA is notable and statistically significant in the 4-yr summary. Combined, these two modifications (QC and TF) generally improve CaPA’s wintertime precipitation analyses with lower bias, partial means closer to observations, higher ETS, lower FAR, and a small increase for POD.

Other modifications are currently being tested to improve CaPA’s wintertime analyses, such as the use of other transfer function, the inclusion of precipitation estimates from weather radar and from IMERG, not currently included in CaPA for the winter season due to its quality control setup. Additional effort should be made to further investigate issues identified in this study, notably regarding the interannual variability of the impact of QC and TF on the results.

Acknowledgments.

The research was supported by Manitoba Hydro (MH) and Mitacs funding. Author Thériault also thanks NSERC Discovery grant and the Canada Research Chairs Program. This research was enabled in part by support provided by Calcul Québec (https://www.calculquebec.ca) and the Digital Research Alliance of Canada (https://alliancecan.ca). We thank Katja Winger for help with using the UQAM servers and Compute Canada facilities. We are grateful to Guy Roy (ECCC) for his help with CaPA. Also, we acknowledge Kevin Sagan, Scott Herbert, and Shane Wruth from Manitoba Hydro. We thank all the MH and ECCC collaborators for their constructive discussions during monthly meetings. We also express our gratitude for the valuable comments provided by the three anonymous reviewers, which have significantly enhanced the quality of this study.

Data availability statement.

All of the experiment files of CTRL, QC, and TF generated from CaPA for this study can be found online (https://doi.org/10.20383/102.0707).

REFERENCES

  • Asong, Z. E., S. Razavi, H. S. Wheater, and J. S. Wong, 2017: Evaluation of Integrated Multisatellite Retrievals for GPM (IMERG) over southern Canada against ground precipitation observations: A preliminary assessment. J. Hydrometeor., 18, 10331050, https://doi.org/10.1175/JHM-D-16-0187.1.

    • Search Google Scholar
    • Export Citation
  • Boluwade, A., T. Stadnyk, V. Fortin, and G. Roy, 2017: Assimilation of precipitation Estimates from the Integrated Multisatellite Retrievals for GPM (IMERG, Early Run) in the Canadian Precipitation Analysis (CaPA). J. Hydrol. Reg. Stud., 14, 1022, https://doi.org/10.1016/j.ejrh.2017.10.005.

    • Search Google Scholar
    • Export Citation
  • CMC, 2018: High Resolution Deterministic Precipitation Analysis System (CaPA-HRDPA). Environment and Climate Change Canada Tech. Note, 16 pp., https://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/tech_notes/technote_capa_hrdpa-450_e.pdf.

  • CMC, 2021: Système d’Analyse Régionale Déterministe de Précipitation (CaPA-ARDP). Environment and Climate Change Canada Tech. Note, 13 pp., https://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/tech_notes/technote_capa_rdpa-520_f.pdf.

  • Colli, M., R. Rasmussen, J. M. Thériault, L. G. Lanza, C. B. Baker, and J. Kochendorfer, 2015: An improved trajectory model to evaluate the collection performance of snow gauges. J. Appl. Meteor. Climatol., 54, 18261836, https://doi.org/10.1175/JAMC-D-15-0035.1.

    • Search Google Scholar
    • Export Citation
  • Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998: The Operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395, https://doi.org/10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Evans, A. M., 2013: Investigation of enhancements to two fundamental components of the statistical interpolation method used by the Canadian Precipitation Analysis (CaPA). M.S. thesis, Dept. of Civil Engineering, University of Manitoba, 307 pp., https://mspace.lib.umanitoba.ca/xmlui/handle/1993/22276.

  • Ferro, C. A. T., and D. B. Stephenson, 2011: Extremal dependence indices: Improved verification measures for deterministic forecasts of rare binary events. Wea. Forecasting, 26, 699713, https://doi.org/10.1175/WAF-D-10-05030.1.

    • Search Google Scholar
    • Export Citation
  • Fortin, V., G. Roy, N. Donaldson, and A. Mahidjiba, 2015: Assimilation of radar quantitative precipitation estimations in the Canadian Precipitation Analysis (CaPA). J. Hydrol., 531, 296307, https://doi.org/10.1016/j.jhydrol.2015.08.003.

    • Search Google Scholar
    • Export Citation
  • Fortin, V., G. Roy, T. Stadnyk, K. Koenig, N. Gasset, and A. Mahidjiba, 2018: Ten years of science based on the Canadian Precipitation Analysis: A CaPA system overview and literature review. Atmos.–Ocean, 56, 178196, https://doi.org/10.1080/07055900.2018.1474728.

    • Search Google Scholar
    • Export Citation
  • Friesen, B., A. Boluwade, P. F. Rasmussen, and V. Fortin, 2017: Assimilation of satellite-based rainfall estimations in the Canadian Precipitation Analysis. J. Hydrol. Eng., 22, 04017049, https://doi.org/10.1061/(ASCE)HE.1943-5584.0001584.

    • Search Google Scholar
    • Export Citation
  • Goodison, B. E., P. Y. T. Louie, and D. Yang, 1998: WMO solid precipitation measurement intercomparison. Instruments and Observing Methods Rep. 67, WMO/TD-872, 212 pp., https://www.wmo.int/pages/prog/www/IMOP/publications/IOM-67-solid-precip/WMOtd872.pdf.

  • Hoover, J., M. E. Earle, P. I. Joe, and P. E. Sullivan, 2021: Unshielded precipitation gauge collection efficiency with wind speed and hydrometeor fall velocity. Hydrol. Earth Syst. Sci., 25, 54735491, https://doi.org/10.5194/hess-25-5473-2021.

    • Search Google Scholar
    • Export Citation
  • Khedhaouiria, D., S. Bélair, V. Fortin, G. Roy, and F. Lespinas, 2020: High-resolution (2.5 km) ensemble precipitation analysis across Canada. J. Hydrometeor., 21, 20232039, https://doi.org/10.1175/JHM-D-19-0282.1.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2017a: Analysis of single-Alter-shielded and unshielded measurements of mixed and solid precipitation from WMO-SPICE. Hydrol. Earth Syst. Sci., 21, 35253542, https://doi.org/10.5194/hess-21-3525-2017.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2017b: The quantification and correction of wind-induced precipitation measurement errors. Hydrol. Earth Syst. Sci., 21, 19731989, https://doi.org/10.5194/hess-21-1973-2017.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2018: Testing and development of transfer functions for weighing precipitation gauges in WMO-SPICE. Hydrol. Earth Syst. Sci., 22, 14371452, https://doi.org/10.5194/hess-22-1437-2018.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2022: How well are we measuring snow post-SPICE? Bull. Amer. Meteor. Soc., 103, E370E388, https://doi.org/10.1175/BAMS-D-20-0228.1.

    • Search Google Scholar
    • Export Citation
  • Leroux, N. R., J. M. Thériault, and R. Rasmussen, 2021: Improvement of snow gauge collection efficiency through a knowledge of solid precipitation fall speed. J. Hydrometeor., 22, 9971006, https://doi.org/10.1175/JHM-D-20-0147.1.

    • Search Google Scholar
    • Export Citation
  • Lespinas, F., V. Fortin, G. Roy, P. Rasmussen, and T. Stadnyk, 2015: Performance evaluation of the Canadian Precipitation Analysis (CaPA). J. Hydrometeor., 16, 20452064, https://doi.org/10.1175/JHM-D-14-0191.1.

    • Search Google Scholar
    • Export Citation
  • Mahfouf, J., B. Brasnett, and S. Gagnon, 2007: A Canadian Precipitation Analysis (CaPA) project: Description and preliminary results. Atmos.–Ocean, 45, 117, https://doi.org/10.3137/ao.v450101.

    • Search Google Scholar
    • Export Citation
  • McTaggart-Cowan, R., and Coauthors, 2019: Modernization of atmospheric physics parameterization in Canadian NWP. J. Adv. Model. Earth Syst., 11, 35933635, https://doi.org/10.1029/2019MS001781.

    • Search Google Scholar
    • Export Citation
  • Mekis, E., N. Donaldson, J. Reid, A. Zucconi, J. Hoover, Q. Li, R. Nitu, and S. Melo, 2018: An overview of surface-based precipitation observations at Environment and Climate Change Canada. Atmos.–Ocean, 56, 7195, https://doi.org/10.1080/07055900.2018.1433627.

    • Search Google Scholar
    • Export Citation
  • Milbrandt, J. A., S. Bélair, M. Faucher, M. Vallée, M. L. Carrera, and A. Glazer, 2016: The Pan-Canadian high resolution (2.5 km) deterministic prediction system. Wea. Forecasting, 31, 17911816, https://doi.org/10.1175/WAF-D-16-0035.1.

    • Search Google Scholar
    • Export Citation
  • Nešpor, V., and B. Sevruk, 1999: Estimation of wind-induced error of rainfall gauge measurements using a numerical simulation. J. Atmos. Oceanic Technol., 16, 450464, https://doi.org/10.1175/1520-0426(1999)016<0450:EOWIEO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Nitu, R., and Coauthors, 2018: WMO Solid Precipitation Intercomparison Experiment (SPICE) (2012–2015). Instruments and Observing Methods Rep. 131, WMO, 1445 pp., https://library.wmo.int/records/item/56317-wmo-solid-precipitation-intercomparison-experiment-spice-2012-2015.

  • Rasmussen, R., and Coauthors, 2012: How well are we measuring snow: The NOAA/FAA/NCAR winter precipitation test bed. Bull. Amer. Meteor. Soc., 93, 811829, https://doi.org/10.1175/BAMS-D-11-00052.1.

    • Search Google Scholar
    • Export Citation
  • Thériault, J. M., R. Rasmussen, K. Ikeda, and S. Landolt, 2012a: Dependence of snow gauge collection efficiency on snowflake characteristics. J. Appl. Meteor. Climatol., 51, 745762, https://doi.org/10.1175/JAMC-D-11-0116.1.

    • Search Google Scholar
    • Export Citation
  • Thériault, J. M., R. E. Stewart, and W. Henson, 2012b: Impacts of terminal velocity on the trajectory of winter precipitation types. Atmos. Res., 116, 116129, https://doi.org/10.1016/j.atmosres.2012.03.008.

    • Search Google Scholar
    • Export Citation
  • Thériault, J. M., R. Rasmussen, E. Petro, J.-Y. Trépanier, M. Colli, and L. G. Lanza, 2015: Impact of wind direction, wind speed, and particle characteristics on the collection efficiency of the double fence intercomparison reference. J. Appl. Meteor. Climatol., 54, 19181930, https://doi.org/10.1175/JAMC-D-15-0034.1.

    • Search Google Scholar
    • Export Citation
  • Wang, X. L., H. Xu, B. Qian, Y. Feng, and E. Mekis, 2017: Adjusted daily rainfall and snowfall data for Canada. Atmos.–Ocean, 55, 155168, https://doi.org/10.1080/07055900.2017.1342163.

    • Search Google Scholar
    • Export Citation
  • Wolff, M. A., K. Isaksen, A. Petersen-Øverleir, K. Ødemark, T. Reitan, and R. Brækkan, 2015: Derivation of a new continuous adjustment function for correcting wind-induced loss of solid precipitation: Results of a Norwegian field study. Hydrol. Earth Syst. Sci., 19, 951967, https://doi.org/10.5194/hess-19-951-2015.

    • Search Google Scholar
    • Export Citation
Save
  • Asong, Z. E., S. Razavi, H. S. Wheater, and J. S. Wong, 2017: Evaluation of Integrated Multisatellite Retrievals for GPM (IMERG) over southern Canada against ground precipitation observations: A preliminary assessment. J. Hydrometeor., 18, 10331050, https://doi.org/10.1175/JHM-D-16-0187.1.

    • Search Google Scholar
    • Export Citation
  • Boluwade, A., T. Stadnyk, V. Fortin, and G. Roy, 2017: Assimilation of precipitation Estimates from the Integrated Multisatellite Retrievals for GPM (IMERG, Early Run) in the Canadian Precipitation Analysis (CaPA). J. Hydrol. Reg. Stud., 14, 1022, https://doi.org/10.1016/j.ejrh.2017.10.005.

    • Search Google Scholar
    • Export Citation
  • CMC, 2018: High Resolution Deterministic Precipitation Analysis System (CaPA-HRDPA). Environment and Climate Change Canada Tech. Note, 16 pp., https://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/tech_notes/technote_capa_hrdpa-450_e.pdf.

  • CMC, 2021: Système d’Analyse Régionale Déterministe de Précipitation (CaPA-ARDP). Environment and Climate Change Canada Tech. Note, 13 pp., https://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/tech_notes/technote_capa_rdpa-520_f.pdf.

  • Colli, M., R. Rasmussen, J. M. Thériault, L. G. Lanza, C. B. Baker, and J. Kochendorfer, 2015: An improved trajectory model to evaluate the collection performance of snow gauges. J. Appl. Meteor. Climatol., 54, 18261836, https://doi.org/10.1175/JAMC-D-15-0035.1.

    • Search Google Scholar
    • Export Citation
  • Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and A. Staniforth, 1998: The Operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 13731395, https://doi.org/10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Evans, A. M., 2013: Investigation of enhancements to two fundamental components of the statistical interpolation method used by the Canadian Precipitation Analysis (CaPA). M.S. thesis, Dept. of Civil Engineering, University of Manitoba, 307 pp., https://mspace.lib.umanitoba.ca/xmlui/handle/1993/22276.

  • Ferro, C. A. T., and D. B. Stephenson, 2011: Extremal dependence indices: Improved verification measures for deterministic forecasts of rare binary events. Wea. Forecasting, 26, 699713, https://doi.org/10.1175/WAF-D-10-05030.1.

    • Search Google Scholar
    • Export Citation
  • Fortin, V., G. Roy, N. Donaldson, and A. Mahidjiba, 2015: Assimilation of radar quantitative precipitation estimations in the Canadian Precipitation Analysis (CaPA). J. Hydrol., 531, 296307, https://doi.org/10.1016/j.jhydrol.2015.08.003.

    • Search Google Scholar
    • Export Citation
  • Fortin, V., G. Roy, T. Stadnyk, K. Koenig, N. Gasset, and A. Mahidjiba, 2018: Ten years of science based on the Canadian Precipitation Analysis: A CaPA system overview and literature review. Atmos.–Ocean, 56, 178196, https://doi.org/10.1080/07055900.2018.1474728.

    • Search Google Scholar
    • Export Citation
  • Friesen, B., A. Boluwade, P. F. Rasmussen, and V. Fortin, 2017: Assimilation of satellite-based rainfall estimations in the Canadian Precipitation Analysis. J. Hydrol. Eng., 22, 04017049, https://doi.org/10.1061/(ASCE)HE.1943-5584.0001584.

    • Search Google Scholar
    • Export Citation
  • Goodison, B. E., P. Y. T. Louie, and D. Yang, 1998: WMO solid precipitation measurement intercomparison. Instruments and Observing Methods Rep. 67, WMO/TD-872, 212 pp., https://www.wmo.int/pages/prog/www/IMOP/publications/IOM-67-solid-precip/WMOtd872.pdf.

  • Hoover, J., M. E. Earle, P. I. Joe, and P. E. Sullivan, 2021: Unshielded precipitation gauge collection efficiency with wind speed and hydrometeor fall velocity. Hydrol. Earth Syst. Sci., 25, 54735491, https://doi.org/10.5194/hess-25-5473-2021.

    • Search Google Scholar
    • Export Citation
  • Khedhaouiria, D., S. Bélair, V. Fortin, G. Roy, and F. Lespinas, 2020: High-resolution (2.5 km) ensemble precipitation analysis across Canada. J. Hydrometeor., 21, 20232039, https://doi.org/10.1175/JHM-D-19-0282.1.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2017a: Analysis of single-Alter-shielded and unshielded measurements of mixed and solid precipitation from WMO-SPICE. Hydrol. Earth Syst. Sci., 21, 35253542, https://doi.org/10.5194/hess-21-3525-2017.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2017b: The quantification and correction of wind-induced precipitation measurement errors. Hydrol. Earth Syst. Sci., 21, 19731989, https://doi.org/10.5194/hess-21-1973-2017.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2018: Testing and development of transfer functions for weighing precipitation gauges in WMO-SPICE. Hydrol. Earth Syst. Sci., 22, 14371452, https://doi.org/10.5194/hess-22-1437-2018.

    • Search Google Scholar
    • Export Citation
  • Kochendorfer, J., and Coauthors, 2022: How well are we measuring snow post-SPICE? Bull. Amer. Meteor. Soc., 103, E370E388, https://doi.org/10.1175/BAMS-D-20-0228.1.

    • Search Google Scholar
    • Export Citation
  • Leroux, N. R., J. M. Thériault, and R. Rasmussen, 2021: Improvement of snow gauge collection efficiency through a knowledge of solid precipitation fall speed. J. Hydrometeor., 22, 9971006, https://doi.org/10.1175/JHM-D-20-0147.1.

    • Search Google Scholar
    • Export Citation
  • Lespinas, F., V. Fortin, G. Roy, P. Rasmussen, and T. Stadnyk, 2015: Performance evaluation of the Canadian Precipitation Analysis (CaPA). J. Hydrometeor., 16, 20452064, https://doi.org/10.1175/JHM-D-14-0191.1.

    • Search Google Scholar
    • Export Citation
  • Mahfouf, J., B. Brasnett, and S. Gagnon, 2007: A Canadian Precipitation Analysis (CaPA) project: Description and preliminary results. Atmos.–Ocean, 45, 117, https://doi.org/10.3137/ao.v450101.

    • Search Google Scholar
    • Export Citation
  • McTaggart-Cowan, R., and Coauthors, 2019: Modernization of atmospheric physics parameterization in Canadian NWP. J. Adv. Model. Earth Syst., 11, 35933635, https://doi.org/10.1029/2019MS001781.

    • Search Google Scholar
    • Export Citation
  • Mekis, E., N. Donaldson, J. Reid, A. Zucconi, J. Hoover, Q. Li, R. Nitu, and S. Melo, 2018: An overview of surface-based precipitation observations at Environment and Climate Change Canada. Atmos.–Ocean, 56, 7195, https://doi.org/10.1080/07055900.2018.1433627.

    • Search Google Scholar
    • Export Citation
  • Milbrandt, J. A., S. Bélair, M. Faucher, M. Vallée, M. L. Carrera, and A. Glazer, 2016: The Pan-Canadian high resolution (2.5 km) deterministic prediction system. Wea. Forecasting, 31, 17911816, https://doi.org/10.1175/WAF-D-16-0035.1.

    • Search Google Scholar
    • Export Citation
  • Nešpor, V., and B. Sevruk, 1999: Estimation of wind-induced error of rainfall gauge measurements using a numerical simulation. J. Atmos. Oceanic Technol., 16, 450464, https://doi.org/10.1175/1520-0426(1999)016<0450:EOWIEO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Nitu, R., and Coauthors, 2018: WMO Solid Precipitation Intercomparison Experiment (SPICE) (2012–2015). Instruments and Observing Methods Rep. 131, WMO, 1445 pp., https://library.wmo.int/records/item/56317-wmo-solid-precipitation-intercomparison-experiment-spice-2012-2015.

  • Rasmussen, R., and Coauthors, 2012: How well are we measuring snow: The NOAA/FAA/NCAR winter precipitation test bed. Bull. Amer. Meteor. Soc., 93, 811829, https://doi.org/10.1175/BAMS-D-11-00052.1.

    • Search Google Scholar
    • Export Citation
  • Thériault, J. M., R. Rasmussen, K. Ikeda, and S. Landolt, 2012a: Dependence of snow gauge collection efficiency on snowflake characteristics. J. Appl. Meteor. Climatol., 51, 745762, https://doi.org/10.1175/JAMC-D-11-0116.1.

    • Search Google Scholar
    • Export Citation
  • Thériault, J. M., R. E. Stewart, and W. Henson, 2012b: Impacts of terminal velocity on the trajectory of winter precipitation types. Atmos. Res., 116, 116129, https://doi.org/10.1016/j.atmosres.2012.03.008.

    • Search Google Scholar
    • Export Citation
  • Thériault, J. M., R. Rasmussen, E. Petro, J.-Y. Trépanier, M. Colli, and L. G. Lanza, 2015: Impact of wind direction, wind speed, and particle characteristics on the collection efficiency of the double fence intercomparison reference. J. Appl. Meteor. Climatol., 54, 19181930, https://doi.org/10.1175/JAMC-D-15-0034.1.

    • Search Google Scholar
    • Export Citation
  • Wang, X. L., H. Xu, B. Qian, Y. Feng, and E. Mekis, 2017: Adjusted daily rainfall and snowfall data for Canada. Atmos.–Ocean, 55, 155168, https://doi.org/10.1080/07055900.2017.1342163.

    • Search Google Scholar
    • Export Citation
  • Wolff, M. A., K. Isaksen, A. Petersen-Øverleir, K. Ødemark, T. Reitan, and R. Brækkan, 2015: Derivation of a new continuous adjustment function for correcting wind-induced loss of solid precipitation: Results of a Norwegian field study. Hydrol. Earth Syst. Sci., 19, 951967, https://doi.org/10.5194/hess-19-951-2015.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Analysis domain of the study (black-outlined frame) together with the location of surface stations considered for assimilation in CaPA (dots), for a specific date (0600 UTC 7 Jan 2019). The colors are the various surface station networks used in this study. For the topography, the colors are the surface covered by plants (green) and the exposed land surface (beige).

  • Fig. 2.

    Number of assimilated observations from automatic stations as a function of 1.5-m wind speed threshold for each winter season (2019–22).

  • Fig. 3.

    Domain of the study and the precipitation analysis at 0600 UTC 7 Jan 2019 produced by the (a) first-guess, (c) CTRL, (e) QC, and (g) TF experiments with the spatial distribution of the assimilated (orange) and rejected (blue) observations. CaPA’s analysis domain for this study is represented by the shaded area. (b) The difference in precipitation amount between CTRL and the first guess. (d) The difference in precipitation amount and the location of observations (stations) assimilated in CTRL and QC. (f) The average value of the wind speed (uv) measured at the stations over the 6-h accumulation period. The difference in precipitation amount between CTRL and QC are the shaded areas in (d) and (e). (h) The difference in precipitation amount between the QC and TF analyses. For (b), (d), (f), and (h), the contour interval is 0.5 mm.

  • Fig. 4.

    The frequency bias index FBI-1 valid for the study domain for CTRL (blue) and QC (red) (a)–(d) for each winter and (e) for the four-winter summary. Here, q is the 6-hourly precipitation accumulation (mm). Filled symbols mean that the differences between CTRL and QC are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

  • Fig. 5.

    As in Fig. 4, but for partial mean.

  • Fig. 6.

    As in Fig. 4, but for equitable threat score (ETS).

  • Fig. 7.

    As in Fig. 4, but for the probability of detection (POD).

  • Fig. 8.

    As in Fig. 4, but for the false alarm rate (FAR).

  • Fig. 9.

    The frequency bias index FBI-1 valid for the study domain for QC (red) and TF (purple) (a)–(d) for each winter and (e) for the four-winter summary. Here, q is the 6-hourly precipitation accumulation (mm). Filled symbols mean that the differences between QC and TF are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

  • Fig. 10.

    As in Fig. 9, but for precipitation partial means.

  • Fig. 11.

    As in Fig. 9, but for ETS.

  • Fig. 12.

    As in Fig. 9, but for POD.

  • Fig. 13.

    As in Fig. 9, but for FAR.

  • Fig. 14.

    Evaluation of CTRL (blue) and TF (purple) for the period of the four winters combined. The y axis is the evaluation metric, and the x axis is the 6-hourly precipitation accumulation threshold (mm). Filled symbols mean that the differences between CTRL and TF are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

  • Fig. 15.

    The frequency bias index FBI-1 valid for the study domain for CTRL (blue) and the first guess (green) (a)–(d) for each winter and (e) for the four-winter summary. Here, q is the 6-hourly precipitation accumulation (mm). Filled symbols mean that the differences between CTRL and QC are statistically significant at the 95% confidence level, based on the bootstrap method; open symbols mean that the differences are not statistically significant.

  • Fig. 16.

    The total precipitation of the background field (the first guess) from HRDPS of each winter.

All Time Past Year Past 30 Days
Abstract Views 131 131 0
Full Text Views 435 435 18
PDF Downloads 333 333 24