Skip to content

Precipitation

In this section, MERRA-2 precipitation is compared to observations reported to the National Center for Environmental Information (NCEI – formerly National Climatic Data Center). Selected surface sites from the NCEI Integrated Surface Database (ISD) files are used for the hourly MERRA-2 comparisons. Global "Summary of the Day" (GSOD) files are used for the comparison to the daily accumulated MERRA-2 precipitation.

Hourly

Hourly Precipitation

Scatter plot of the MERRA-2 hourly precipitation vs. the hourly station observations values from the selected NCEI ISD files for the years 2001 - 2019.

Hourly Precipitation

The total number of hourly pairs is 37,282,572.

Daily Accumulations

Accumulated Days Slope Intercept Rsqrd Bias RMSE
1 0.36 2.06 0.15 0.80 7.96
5 0.57 6.12 0.45 2.15 15.12
15 0.53 16.86 0.53 6.35 31.35
30 0.63 32.98 0.54 12.68 53.52
1 Day Accumulated Precipitation

Scatter plot of the re-gridded MERRA-2 1 day accumulated precipitation vs. the daily station observations values from the NCEI GSOD files for every 3rd year from 1981 – 2014.

Accumulated Precipitation 1

5 Day Accumulated Precipitation

Scatter plot of the re-gridded MERRA-2 5 day accumulated precipitation vs. the daily station observations values from the NCEI GSOD files for every 3rd year from 1981 – 2014.

Accumulated Precipitation 5

15 Day Accumulated Precipitation

Scatter plot of the re-gridded MERRA-2 15 day accumulated precipitation vs. the daily station observations values from the NCEI GSOD files for every 3rd year from 1981 – 2014.

Accumulated Precipitation 15

30 Day Accumulated Precipitation

Scatter plot of the re-gridded MERRA-2 30 day accumulated precipitation vs. the daily station observations values from the NCEI GSOD files for every 3rd year from 1981 – 2014.

Accumulated Precipitation 30

Scatter Plot Notes

  • The legend in the upper left corner of each scatter plot gives the associated mean Bias and RMSE along with the Slope, Intercept, and Pearson correlation coefficient associated with the linear least squares fit to the plotted values (black line).
  • The color bar along the right side of the scatter plot provides a measure of the frequency of the NCEI, MERRA-2 precipitation data pairs per bin in plot (5 mm bins for day accumulations).

Contingency Table

Many methods exist when it comes to the validation of precipitation (Yates, 2006; Ebert, 2007). Density scatter plots are a useful way to describe and display the validation statistics. However, the density scatter plots of the precipitation validation, due to the random nature of the daily precipitation, are dominated by days with little or no precipitation while days with heavy precipitation are less frequent. Categorical verification, in this case contingency tables, offer another way to examine the daily precipitation data. What the contingency table shows are the cases when the MERRA-2 daily precipitation and ground site daily precipitation:

  1. Measured no precipitation on the same days ("correct negative").
  2. Measured precipitation on the same days ("hits"; regardless of amount).
  3. MERRA-2 measured precipitation while ground site does not ("false alarms").
  4. MERRA-2 measured no precipitation while ground site did ("misses").

For this contingency table analysis of the MERRA-2 daily precipitation, all the days when the ground site/MERRA-2 data pairs measured a valid precipitation amount, including days with no precipitation, were counted for every 3rd year during the 1981-2014 period. These "data days" total 8,958,351 days. The top table in the Hits Contingency Table shows a breakdown of the four categories mentioned previously, with the MERRA-2 data presented by the "rows" on the left side of the table and the NCEI ground site data presented by the "columns" on the top side of the table. Row 1 and Column 1 depict the "No Precipitation" category for the MERRA-2 and NCEI ground site data, respectively. Where Row 1 and Column 1 intersect shows the total number of data days (4,079,063) where both data sets measured no precipitation on the same days ("correct negative"). This represented 45.53% of the total data days counted.

Row 2 and Column 2 depict the "Precipitation" categories for the MERRA-2 and NCEI ground site data, respectively. Their intersection shows the total number of data days (1,858,936) where both data sets measured precipitation regardless of amount on the same days ("hits"). This represented 20.75% of the total data days counted. The "hits" category is broken down further in sub-categories, depicted in the lower table in the Hits Contingency Table, along with statistics. The sub-categories consist of data days when the MERRA-2 precipitation is less than the NCEI ground site, when the MERRA-2 precipitation is greater than the NCEI ground site, and when the MERRA-2 precipitation is equal to the NCEI ground site. By "equal" means within the measurement accuracy of the NCEI ground site instruments (0.508 mm). Out of the total number of data day "hits", 248,069 (13.36%) data days occur when the MERRA-2 precipitation is equal to the NCEI ground site precipitation. Both the MERRA-2 precipitation and NCEI ground site precipitation share similar averages, standard deviations, and maximum values, though the minimum values differ by 0.50 mm. The MERRA-2 precipitation is less than the NCEI ground site precipitation on 903,183 (48.59%) data days out of the total number of data day "hits". As expected, the MERRA-2 precipitation average and standard deviations are roughly 2.5 times less than the NCEI ground site precipitation average and standard deviations. The NCEI ground site precipitation maximum and minimum values are roughly 2 and 4 times greater than the MERRA-2 precipitation maximum and minimum values, respectively. The MERRA-2 precipitation is more than the NCEI ground site precipitation on 707,684 (38.07%) data days out of the total number of data day "hits". As expected, the MERRA-2 precipitation average and standard deviations are roughly 2 times more than the NCEI ground site precipitation average and standard deviations. The MERRA-2 precipitation maximum and minimum values are roughly 1.5 times greater than the NCEI ground site precipitation maximum and minimum values.

Hits Contingency Table

Contingency table with focus on the "Hits" category between the re-gridded MERRA-2 precipitation and values based upon NCEI GSOD data for every 3rd year from 1981 – 2014.

Contingency Table

Row 1 and Column 2 depict the "No Precipitation" and "Precipitation" categories for the MERRA-2 and NCEI ground site data, respectively. Their intersection shows the total number of data days (200,580) where the MERRA-2 precipitation data measured no precipitation while the NCEI ground site precipitation data measured precipitation, regardless of amount, on the same days ("misses"). This represented 2.24% of the total data days counted. The "misses" category is broken down further, depicted in the lower table in the Misses Contingency Table, along with statistics. Shown are the data days when the MERRA-2 precipitation has no precipitation and the NCEI ground site has a precipitation amount greater than the measurement accuracy of the NCEI ground site instruments ( 0.508 mm). In these cases, the NCEI ground site precipitation has an average of 4.83 mm, a standard deviation of 9.83 mm, a maximum value of 410.97 mm, and a minimum value of 0.762 mm.

Misses Contingency Table

Contingency table with focus on the "Misses" category between the re-gridded MERRA-2 precipitation and values based upon NCEI GSOD data for every 3rd year from 1981 – 2014.

Contingency Table

Row 2 and Column 1 depict the "Precipitation" and "No Precipitation" categories for the MERRA-2 and NCEI ground site data, respectively. Their intersection shows the total number of data days (2,819,772) where the MERRA-2 precipitation data measured precipitation while the NCEI ground site precipitation data measured no precipitation on the same days ("false alarms"). This represented 31.48% of the total data days counted. The "false alarm" category is broken down further, depicted in the lower table in the False Alarms Contingency Table, along with statistics. Shown are the data days when the MERRA-2 precipitation measured precipitation and the NCEI ground site measured a precipitation amount less than and equal to the measurement accuracy of the NCEI ground site instruments, ( 0.508 mm). In these cases, the MERRA-2 precipitation has an average of 3.02 mm, a standard deviation of 5.02 mm, a maximum value of 213.31 mm, and a minimum value of 0.26 mm.

False Alarms Contingency Table

Contingency table with focus on the "False Alarms" category between the re-gridded MERRA-2 precipitation and values based upon NCEI GSOD data for every 3rd year from 1981 – 2014.

Contingency Table

The findings from the contingency tables show additional information not easily seen in the density scatter plots in the Precipitation Overview. On data days when the MERRA-2 precipitation measured no precipitation, it agreed with the NCEI ground site precipitation 45.53% of the time ("correct negatives"), and only disagreed 2.24% of the time ("misses"). The percent of "misses" is relatively low compared to the other three categories in the table, with the NCEI ground site precipitation measurement average only being 4.83 mm or 0.19 inches. On data days when the MERRA-2 precipitation measured precipitation, the NCEI ground site also measured precipitation 20.75% of the time ("hits"), but disagreed 31.48% of the time ("false alarms"). On data days when the NCEI ground site measures precipitation, it’s encouraging to note MERRA-2 predicts precipitation (20.75%) on the same data days more often than it doesn’t (2.24%). In the "hits" category, however, only 2.77% of the total data days the MERRA-2 and NCEI ground site precipitation agree exactly within the accuracy of the ground site instrument (0.508mm). The percent of "false alarms" represents roughly one-third of data days in the table, with the MERRA-2 precipitation measurement average only being 3.02 mm or 0.12 inches. MERRA-2 is more likely to predict measured precipitation when none occurs at a ground site than predict no measured precipitation when it does occur at a ground site.

Localized Comparison

The scatter plots below provide a global perspective covering a wide range of precipitation as well as ground station with varying levels of quality control. The ability of the MERRA-2 to provide precipitation data for local applications can be assessed in the plots where the daily averaged MERRA-2 precipitation associated with the half-degree cell covering the Houston, TX region is plotted against the corresponding daily averaged values from surface measurements at the Houston Hobby airport. The plots cover the time period 1981 – 2017. Note, here again as in the data from a global perspective, the agreement improves as the averaging period increased from 1-day through 30-days, and as with the data in the Accumulated Precipitation Scatter Plots (below) the improvement is marginal after 5-days. The data points circled in red are from the time period that Hurricane Harvey passed over the Houston area.

Success

The localized day based accumulated precipitation time series plot show high correlation between MERRA-2 and the surface observations. Similar results to those in below were also observed at other surface stations in Texas and in the Louisiana.

Localized 1 Day Accumulated Precipitation

Time series plot of the re-gridded MERRA-2 1 day accumulated precipitation vs. the daily station observations form William P. Hobby Airport, Houston, TX from 1981 – 2017.

Daily Mean All-Sky SW Insolation

Localized 5 Day Accumulated Precipitation

Time series plot of the re-gridded MERRA-2 5 day accumulated precipitation vs. the daily station observations form William P. Hobby Airport, Houston, TX from 1981 – 2017.

Daily Mean All-Sky SW Insolation

Localized 10 Day Accumulated Precipitation

Time series plot of the re-gridded MERRA-2 10 day accumulated precipitation vs. the daily station observations form William P. Hobby Airport, Houston, TX from 1981 – 2017.

Daily Mean All-Sky SW Insolation

Localized 15 Day Accumulated Precipitation

Time series plot of the re-gridded MERRA-2 15 day accumulated precipitation vs. the daily station observations form William P. Hobby Airport, Houston, TX from 1981 – 2017.

Daily Mean All-Sky SW Insolation

Localized 30 Day Accumulated Precipitation

Time series plot of the re-gridded MERRA-2 30 day accumulated precipitation vs. the daily station observations form William P. Hobby Airport, Houston, TX from 1981 – 2017.

Daily Mean All-Sky SW Insolation

Localized Hurricane Harvey

Time series plot plot of the 1-day average precipitation cover Hurricane Harvey (i.e. August 1, 2017 to September 30, 2017)

Daily Mean All-Sky SW Insolation