Introduction

In this document I will explore how to create the first part of the evaluation system I proposed. The working title of this is the “Forecast-Hour Evaluation.” The idea here is that we are looking at the performance of the model by looking at how it performed with different start times (using the most recent 00-hr forecast as input).

## Warning in rgl.init(initValue, onlyNULL): RGL: unable to open X11 display
## Warning: 'rgl.init' failed, running with 'rgl.useNULL = TRUE'.

Read Model and Observation Data

Read WRF Data

For this evaluation system we need to look at three different output folders. Here we use the folders named, forecast_day_minus_0, forecast_day_minus_1, forecast_day_minus_2. The contents of each of these folders will be similar: wrfout files for 86 forecast hours and time-series data for different locations of interest. Here we will first read the forecast data.

Read the OBS Data

Now we will read the observation data from the ASOS stations. The script that downloads the data is in ./obs_station_day_minus_0/dl_ny_asos.py. The lines for the dates to download need to be changed before running it. Once the files are download, the lines below reads the data and adds column names.

Unit Conversion

Model and observation data do not share the same units for the same variable. For temperature, WRF is in Kelvin and ASOS is in degreesF. For winds, WRF is in m/s and ASOS is in knots. The formulas used to convert the numbers to a common system is shown here. For temperature I will use Kelvin, and m/s for wind speeds.

Combined Data Frames

Now we have one data frame for all the observations, and three (3) data frames of the WRF data (one data frame per forecast init time). The lines below provide a visual of the data frames.

Locations for Plots

Match Times for Model and Observations

Select Day of Interest

Time-matching is performed using a routine that can be found in Analysis01-Time_Matching_Problem.Rmd. The time matching will be done per variable. For the Forecast-Hour Evaluation product, we will focus on the temperature, wind speed and wind direction variables. Also, now that we have read all the TS data and ASOS data, we need to extract the day of interest, or doi for the time-series.

Note that for this product the “day of interest” will always be the UTC date of the day before.

We now have filtered data frames for the observations and model data for the day of interest.

Next, we will select only the temperature data for comparing the model and observations. This needs to be done on a per station basis. Note that we use the function drop_na() to drop rows which contain NaN or NA data. Since each variable is measured at different intervals, not all variables will have data available at every time step in the ASOS data. The functions may be too sensitive to missing data and thus we take care to remvove it here from the observations, after we have isolated a particular variable.

Temperature Time-Matching

Location: JFK

Location: LGA

Location: NYC

Wind Speed Time-Matching

Location: JFK

Location: LGA

Location: NYC

Wind Direction Time-Matching

Location: JFK

Location: LGA

Location: NYC

Forecast Hour Evaluation for JFK

For the temperature data I will use Bias, RMSE and MAE for the comparison statistics

JFK - WRF 2-m Temperature (K) Performance
Forecast.Init BIAS RMSE MAE
1 WRF D-0 -0.438 2.000 1.749
2 WRF D-1 -0.314 1.721 1.500
3 WRF D-2 -1.259 1.962 1.713
JFK - WRF 10-m Wind Speed (m/s) Performance
Forecast.Init BIAS RMSE MAE
1 WRF D-0 -1.396 2.503 2.129
2 WRF D-1 -1.071 2.239 1.969
3 WRF D-2 -0.517 1.810 1.491
JFK - WRF 10-m Wind Direction (degN) Performance
Forecast.Init RMSE MAE
1 WRF D-0 26.737 21.800
2 WRF D-1 36.694 29.353
3 WRF D-2 31.000 24.278

Forecast Hour Evaluation for LGA

LGA - WRF 2-m Temperature (K) Performance
Forecast.Init BIAS RMSE MAE
1 WRF D-0 -0.618 1.665 1.466
2 WRF D-1 -1.024 1.960 1.690
3 WRF D-2 -1.928 2.442 2.078
LGA - WRF 10-m Wind Speed (m/s) Performance
Forecast.Init BIAS RMSE MAE
1 WRF D-0 -2.136 2.627 2.298
2 WRF D-1 -2.097 2.469 2.201
3 WRF D-2 -1.783 2.234 1.957
LGA - WRF 10-m Wind Direction (degN) Performance
Forecast.Init RMSE MAE
1 WRF D-0 22.188 17.448
2 WRF D-1 26.720 19.628
3 WRF D-2 23.155 17.343

Forecast Hour Evaluation for NYC

NYC - WRF 2-m Temperature (K) Performance
Forecast.Init BIAS RMSE MAE
1 WRF D-0 1.502 2.011 1.629
2 WRF D-1 1.000 1.827 1.384
3 WRF D-2 0.109 1.360 1.199
NYC - WRF 10-m Wind Speed (m/s) Performance
Forecast.Init BIAS RMSE MAE
1 WRF D-0 -1.337 1.621 1.501
2 WRF D-1 -1.332 1.627 1.472
3 WRF D-2 -1.242 1.532 1.411
NYC - WRF 10-m Wind Direction (degN) Performance
Forecast.Init RMSE MAE
1 WRF D-0 39.795 37.173
2 WRF D-1 43.469 40.224
3 WRF D-2 43.918 41.799

uWRF and ASOS Time-Series Visualization

Temperature

## Warning: Removed 288 rows containing missing values (geom_point).
## Warning: Removed 288 rows containing missing values (geom_path).

Wind Speed

Wind Direction