Drought Forecast Verifications

How to read these maps

Initial Condition Map

• The initial condition is obtained from the drought nowcast system. It is the daily average soil moisture expressed as precentiles with respect to the historical distribution.
• The percentile value can be interpreted as the probability of historical observations being smaller than the current value.
• Visually, the green regions are wetter than normal and yellow/red regions are drier than normal.
• The contours indicate the change in soil moisture in the last 7 days, also expressed as changes in percentiles.

• These forecasts are produced monthly at the beginning of each month using the initial condition that is closest to the first day of the month.
• The CFS-base forecast uses all CFS seasonal forecast initialized during the previous month. For example, to make seasonal drought forecast starting from 20080801, we use CFS forecast initialized during 20080701 to 20080731 (62 members) to form the forecast distribution before applying the Bayesian merging method.
• The CPC outlook-based forecast uses the CPC output releases during the previous month as the forecast distribution for each climate division. For example, to make seasonal drought forecast starting from 20080801, we use CPC outlook released on 20080717.
• Each column is the 6-month soil moisture forecast from one forecast approach, and each row is the forecasts from all approaches for each individual month, showing monthly average.
• Since we are doing ensemble forecast, the ensemble median is shaded and the ensemble spread is contoured. The color scheme is the same as in the initial condition.
• The ensemble spread is expressed as the interquartile range of the ensemble, that is the distance between the 75th percentile and the 25th percentile of the ensemble, and the distance is expressed as difference between the to soil moisture percentiles in the historical distribution.
• The first column (CFS-based forecast, full) use all the 20 ensemble members to calculate the median and spread, and the second column (CFS-based forecast, most likely) uses only the top 7 members out of the 20. Ten of the twenty members are selected based on a historical-analog criterion, hence they can be sorted. The top 7 members are the most similar ones to the forecast condition.

Target Month	CFS-based Forecast (full) released on 200908	CFS-based Forecast (most likely) released on 200908	CPC Outlook-based Forecast released on 200908	Nowcast (monthly mean)
Initial condition
200908
200909
200910
200911
200912
201001