07.02.2013 Views

Fourth Study Conference on BALTEX Scala Cinema Gudhjem

Fourth Study Conference on BALTEX Scala Cinema Gudhjem

Fourth Study Conference on BALTEX Scala Cinema Gudhjem

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Multichannel Microwave Radiometer flown <strong>on</strong> Nimbus-7).<br />

A linear regressi<strong>on</strong> yields:<br />

SM1 = -1.32 t' + 21.52 rr'' - 1.341 tb10h + 5.5 (2)<br />

where t' denotes the anomaly against the l<strong>on</strong>g-time local air<br />

temperature averaged over the last 3 m<strong>on</strong>th. rr'' is the<br />

anomaly against the l<strong>on</strong>g-term local annual cycle of rainfall<br />

averaged over the past two m<strong>on</strong>ths, and tb10h the<br />

horiz<strong>on</strong>tally polarized brightness temperature from satellite.<br />

A correlati<strong>on</strong> of 0.609 is attained which might seem to be<br />

relatively weak. But we have to keep in mind that the<br />

remaining part of the variance (about 15% of the total<br />

variance) is c<strong>on</strong>sidered here which includes nearly the<br />

complete error variance of the soil moisture observati<strong>on</strong>s.<br />

Hence, it is more difficult to explain a larger part of it,<br />

(which is <strong>on</strong> the other hand small compared to the total<br />

variance).<br />

Thus, soil moisture can be derived with a correlati<strong>on</strong> of<br />

nearly 0.8 by combining eq.(1) and eq.(2). It is obvious that<br />

about three quarters of the spatial variance is explained in<br />

this way, as it is treated explicitly in equati<strong>on</strong> and figure 1.<br />

For the temporal variance the circumstances are more<br />

complicated since observati<strong>on</strong> errors, annual cycle and interannual<br />

variability are lumped together. To asses the<br />

performance of the algorithm we computed the mean<br />

m<strong>on</strong>thly soil moisture for the entire regi<strong>on</strong> and compared the<br />

results to the measurements. The mean annual cycle is<br />

reproduced with an excellent correlati<strong>on</strong> of 0.979. An<br />

analogous procedure provides for the inter-annual variability<br />

a value of 0.525.<br />

4. Verificati<strong>on</strong><br />

In order to assess the algorithm's quality, it has to be applied<br />

to independent data. Spread over Illinois, 19 soil moisture<br />

stati<strong>on</strong>s operate since several decades. We extracted<br />

measurements from the period 1979 to 1999 and computed<br />

the l<strong>on</strong>g-time mean soil moisture in the uppermost meter at<br />

each stati<strong>on</strong>. Such temporal averages are alternatively<br />

derived with our proposed retrieval algorithm by using<br />

globally available informati<strong>on</strong> of four climatological<br />

parameters as given in eq.(1). An excellent agreement<br />

between algorithm and measurement is found. The measured<br />

total mean of 330 mm is reproduced with a deviati<strong>on</strong> of <strong>on</strong>ly<br />

4 mm. The accomplished quality c<strong>on</strong>trol is quite an acid test<br />

since the algorithm is transferred to another c<strong>on</strong>tinent and<br />

into a climate regi<strong>on</strong> with soil moistures much higher than<br />

those prevailing in the former Soviet Uni<strong>on</strong> where the<br />

algorithm has been derived.<br />

However, at first glance the correlati<strong>on</strong> between<br />

measurements and retrieval appears with 0.235 rather low.<br />

The explanati<strong>on</strong> becomes obvious when we compare the<br />

error variance of the algorithm with the variability that is<br />

covered by the Illinois measurements. The error variance of<br />

the algorithm appears as unexplained variance in figure 1<br />

and amounts to about 51 mm 2 which is even larger than the<br />

variability comprised in the Illinois dataset. Thus,<br />

c<strong>on</strong>sidered <strong>on</strong> c<strong>on</strong>tinental scale all 19 stati<strong>on</strong>s in Illinois are<br />

located in immediate mutual proximity, and represent<br />

effectively <strong>on</strong>ly a single site. Hence, the detected low<br />

correlati<strong>on</strong> is a foreg<strong>on</strong>e c<strong>on</strong>clusi<strong>on</strong>.<br />

5. Applicati<strong>on</strong><br />

In the following, the presented algorithm is used to validate<br />

the soil moisture of REMO (REgi<strong>on</strong>al MOdel). A ten-year<br />

climate run provided by the MPI for Meteorology in<br />

Hamburg is examined in two aspects according to the twostep<br />

retrieval algorithm. The first is c<strong>on</strong>cerned with l<strong>on</strong>g-<br />

- 13 -<br />

Figure 3. L<strong>on</strong>g-time mean soil moisture from<br />

REMO (right) and from the retrieval (left).<br />

term averages, the sec<strong>on</strong>d is addressed to the temporal<br />

variance. As an exemplary regi<strong>on</strong> the Oder catchment is<br />

analysed. The modelled l<strong>on</strong>g-time mean soil moisture<br />

averaged over the entire model area is almost identical to<br />

the corresp<strong>on</strong>ding total mean retrieved by the algorithm.<br />

However, str<strong>on</strong>g regi<strong>on</strong>al differences occur especially near<br />

the border between Sweden and Finland where the<br />

prescribed model soil type cause extremely high soil<br />

moistures. If true at all, such extreme values are not<br />

resolvable by the algorithm, because the used datasets<br />

have a coarser resoluti<strong>on</strong> than the model. For the Oder<br />

catchment the temporal evoluti<strong>on</strong> of soil moisture is<br />

computed as it is modelled by REMO and as it is retrieved<br />

by our algorithm. The model shows in general a more<br />

pr<strong>on</strong>ounced annual cycle and inter-annual variati<strong>on</strong>s found<br />

in the model are not reflected in the retrieval. However,<br />

the total averages are again in good agreement showing<br />

that no regi<strong>on</strong>al bias is found for this particular catchment.<br />

6. C<strong>on</strong>clusi<strong>on</strong>s<br />

The major part of soil moisture variance originates from<br />

spatial differences between l<strong>on</strong>g-time means at each<br />

locati<strong>on</strong>. Our algorithm is capable to reproduce this<br />

variance to a large extend by using easily available data,<br />

i.e. precipitati<strong>on</strong>, vegetati<strong>on</strong> density, soil texture and<br />

terrain slope. In a sec<strong>on</strong>d step the temporal variance is<br />

explained so that the retrieved annual cycle is found to be<br />

in good agreement with the measurements. Independent<br />

soil moisture measurements from Illinois c<strong>on</strong>firm the<br />

quality of the presented retrieval algorithm.<br />

Using the retrieval as validati<strong>on</strong> for modelled soil moisture<br />

from REMO, large differences occur in both, the regi<strong>on</strong>al<br />

l<strong>on</strong>g-term mean soil moisture and the temporal variance<br />

within a regi<strong>on</strong>. However, total averages are in good<br />

agreement.<br />

References<br />

Vinnikov, K. Ya. & I. B. Yeserkepova, Soil moisture:<br />

empirical data and model results. J. Climate, 4, pp. 66-79,<br />

1991.<br />

Willmott, C. J. & S. M. Robes<strong>on</strong>, Climatologically Aided<br />

Interpolati<strong>on</strong> (CAI) of Terrestrial Air Temperature.<br />

Internati<strong>on</strong>al Journal of Climatology, 15, pp. 221-229,<br />

1995.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!