The verifications are based on changes in the Drought Monitor between the start and the end of the Drought Outlook period.
The Drought Outlook verifications are based on the difference between (1) the current Drought Monitor when the Drought Outlook was released, and (2) the current Drought
Monitor for the morning after the last day of the Drought Outlook period.
The Drought Monitor (DM) is issued each Thursday morning, valid two days earlier (Tuesday, 8 a.m. Eastern time), so the DMs serving as bases for the initial and final drought
conditions (verification period) are normally: (1) Tuesday, 8 a.m. Eastern before the Drought Outlook (DO) release, and (2) Tuesday, 8 a.m. Eastern before the last day of
the final month of the DO period. If the 1st of the month is a Tuesday after the valid DO period, we use that morning's DM since it only considers 4 to 8 hours of the
ensuing month. The dates of the DMs used in the verifications are indicated on the graphic.
The Seasonal Drought Outlook (SDO) takes effect the day of its release (third Thursday of the month) and NOT at the beginning of the next month as is the case with the CPC
Seasonal Temperature and Precipitation Outlooks and initial Monthly Temperature and precipitation Outlooks (all released concurrently). The Monthly Drought Outlook (MDO)
is released on the last day of the prior month and uses the most recent Drought Monitor as a start. The MDO is released concurrently with the updated Monthly Temperature
and Precipitation Outlooks.
The Drought Monitor validation maps are smoothed and interpolated.
The Drought Monitor (DM) is high-resolution analysis, but the Drought Outlook (DO) is intended to forecast trends over broad areas. Therefore, the DMs used for DO
verification are initially smoothed and interpolated. The DM (which assesses drought intensity at any location as an integer between 0 and 4, where 0 = no drought/D0, and
4 = D4), is resolved to a 4 km grid covering the contiguous 48 States [484,602 grid points]. The value at each of those points is adjusted based on the values of all grid
points within a 48 km radius, with the influence of each grid point declining only slightly with distance from the point being recalculated. The result is a 4 km grid of
values that is smoother than the initial DM, and interpolated to decimal values rather than limited to integers between 0 and 4.
Using these smoothed and interpolated DM grids, a DM change grid is derived for the DO period by subtracting the initial DM value from the ending value at each grid point.
Presently, DM change grids are not calculated in Alaska, Hawaii, or Puerto Rico, and the DOs are not verified in these regions.
The difference between the starting and ending Drought Monitor analyses is compared to the Drought Outlook to determine the Outlook's accuracy.
At each of the grid points, the difference between the smoothed and interpolated DMs at the start and end of the DO period is compared, and each point is scored as such:
- If there was (a) no drought at the start of the period, and (b) no drought at the end of the period, and (c) the DO indicated that drought was not expected to
develop, the point is not scored and is ignored in all calculations. No drought is considered abnormally dry (D0) or wetter, drought is considered D1 or drier.
- If the DO was for improvement or removal, the point is scored a "hit" [correct] where the smoothed and interpolated DM value was at least 0.5 [half a category]
better at the end of the period than it was at the beginning. Otherwise, a forecast for improvement or removal is a "miss" [incorrect].
- If the DO was for persistence, the point is scored a "miss" where the smoothed and interpolated DM value was at least 0.5 better at the end of the period than
it was at the beginning. Otherwise, a forecast for persistence is a "hit".
- If the DO was for drought development, the point is scored a "hit" if there was drought depicted by the ending DM, and a "miss" if there was not.
- At locations where there was no drought to start but where the ending DM indicated drought, a "miss" is scored if there was no forecast for development.
The maps on each verification graphic use outlines to depict what the DO called for, and blue and red shadings to show areas that were "hit" [correct] and areas that were
"missed" [incorrect], respectively.
In the tables below the maps, counts of the number of grid points "hit" and the number "missed" are identified by forecast classification.
Verification "raw scores" (or "scores") indicate what percent of the grid points being scored were "hits" - (# of hits)/((# of hits) + (# of misses)). Raw
scores can range from 0% to 100%.
Scores (more precisely, "raw scores") are determined by what percentage of the points that were "hit" or "missed" were, in fact, "hit." The national verification "raw
score" is shown in the table below the map, along with a count of "hit" and "missed" grid points.
Verification "SKILL" compares the "raw score" for the Drought Outlook verification to the "raw score" for a baseline forecast of "persistence" with no areas of
"development." Skill can range from -100 to +100, and positive skill indicates that the Drought Outlook scored better than the baseline.
After a national raw score is determined for the DO, the "skill" for the Outlook is determined by comparing the raw score to how a forecast of drought persistence with no
drought development would have scored. This hypothetical baseline forecast is scored in the same manner as the actual DO described earlier. The baseline score is then
subtracted from the actual DO score, and the result is defined as the "skill." Theoretically, it could range from -100 to +100 (if the [actual]/[baseline] score was 0 and
the [baseline]/[actual] score was 100, respectively).
Verification graphics, which include a map of hits and misses and a table containing: (a) grid point counts of hits and misses by forecast type and for the contiguous 48
States as a whole; (b) the national verification raw score; (c) grid point counts of hits and misses for a baseline forecast of drought persistence with no drought
development; (d) the baseline hypothetical raw score, and (e) the verification skill. This type of verification is available for the Seasonal Drought Outlooks since the
updated ND09J10 release (Nov. 5, 2009), and for the Monthly Drought Outlooks since their first release on June 30, 2013. A simple visual verification (non-statistical, no
scores) of the SDOs were made for the SON06 to initial ND09J10 outlooks, and no verifications are available prior to the SON06 outlook.
|