‹header›
‹date/time›
Click to edit Master text styles
Second level
Third level
Fourth level
Fifth level
‹footer›
‹#›
Precipitation is an intermittent process
SD was a fraction of rainfall for rain events
SD was 1mm/h * mean annual precip/average mean annual precip when N(0,1) > 3
Wind, downward radiation and precip less than zero truncated to zero.
Histogram of percentage of catchments with a given error
Model is already fairly dry so -> wet bias to get error in initial condition and forecasts
Wet bias in precip -> wet bias in model
Assimilation -> dry bias to counteract
Bias improved in both but rms error greater for 4%
Weighted average of all catchments for entire year from daily (soil moist) and 10 day averages (evap)
Obs error <3%v/v required for soil moisture rms to be better than no assim
If perfect model error covariance statistics assimilation should not be worse than no assimilation – biased by high rms for catch with deep soil?
Improvement in moisture bias up to 5.5%v/v
Profile soil moisture is dry biased for 0% error assim –- result of violation of key assumption in the Kalman filter that model error is unbiased (zero mean Gaussian white noise); tries to counteract continual wet bias in surface soil moisture forecast by drying deep soil due to linkage with deep soil moisture –- as obs error is increased assimilation has less of an impact an hence returns to wet bias. Without assim evap is +ve biased because soil moisture is biased wet from bias in precip – with assim some of this bias is removed May also be a result of errors in atmospheric forcing (ie correct soil moisture does not guarantee correct evap) Also, extra rain during inter-obs is available for evap; evap data integrated over entire window (not instantaneous)
Improvement in evap bias though up to 3%v/v!
Why does evap rms increase above no assimilation for all runs even though soil moisture is improved?
Yearly average for individual catchments from 10 day averages
Few small catchments with large rms errors skew the results
Bias also effects rms error
Some similarities between no assim and assim (particularly Alaska region)
Large +ve bias in west and small –ve bias in east
Why wet and dry biases? -> soil moisture
Yearly average for individual catchments from daily values
Rms error largely a result of bias
Bias in “no assim” largely from initial condition; particularly in the north
Dry bias is the result of wet bias in precip –- see next slide also
Wet bias due to dry soil (at wilting) – wet obs makes soil wetter but dry obs cannot make soil drier as limited by wilting point -> effectively a wet bias in the observations as wet and dry errors don’t cancel out. Becomes more severe for greater errors. Some is a result of deep soil – surface less connected with profile for soil depths greater than 3m.
Compare with previous slide
Large precip bias mainly south east -> dry soil bias
Deep soil central N America -> surface/deep soil disconnect
Dry -> wet bias in assim as truncated to wilting
Wet bias and high evaporative demand -> strong +ve bias in evap
Further illustration of biased forcing data
Compare with next slide (from before)
Increased wet bias in no assim and increased dry bias in assim – particularly during summer
Comparison with previous slide only
Compare with previous 2 slides
No precip gives dry bias in no assim (not just summer) and wet bias in assim – worse than for wet bias in precip -> leads to comparison in next slide
Rms errors in soil moisture and evap decreased irrespective of precip bias when assimilation but best results when precip is unbiased
Moisture bias in assimilation is inversely related to precip bias
Surface soil moisture and evap bias largely unaffected by precip bias 
Dry precip bias -> dry evap bias and visa verca
Unbiased precip -> unbiased evap? Assimilation or not