Soon 8 candles for SMOS!!!!!! (6/8)

Category : CATDS, L2, L3, L4, Model

After the illustrations of some striking results over oceans, we can only marvel, especially as many other aspects were not covered.  Eight years ago we did not have any of such applications and science return. Those span from rainfall estimates over oceans to wind speed retrievals for strong winds (tropical storms, hurricanes and the like) where wind scatterometers do saturate for lower wind speeds. SMOS, Aquarius and now SMAP do show that L band measurements bring forward many new science obviously but also many very practical and societal applications which are not fulfilled without them.


Caption: IRMA (2017 09 07) as seen from SMOS in terms of surface wind speed (N. Reul)

This also applies for land of course where new applications blossomed at an unprecedented rate.

It exemplifies, to me at least, how real measurements can never be replaced by proxies. The first radar for EO flew in 1977 (yes 40 years ago!), the scatterometers with Envisat have been available since 1991 but we have yet to see a real soil moisture map from these. Intrinsically active systems are more sensitive to structure that to content and radar soil moisture are at best validated only over small areas where all is known, and similarly to scatterometers, rely on change detection (yes I know I am partial but I can claim that I started fiddling with radars 40 years ago and was one of the pro SCAT over land (convincing ESA to make the sigma nought triplets available over land which was not originally planned incidentally), but to realise soon that it was no game for absolute retrievals). Which means that they have to be scaled and that the validity at point (xi,yi) and no relationship with the validity at point (xj, yj) etc … but this is another story…To make a long story short a nicely coloured map has never make an accurate map.

With L band radiometry no such issues and if properly done, you have access to the soil moisture per se. As a direct consequence, and in opposition to active systems, a few months only after the release of the data the first applications emerged. We saw the first use in food security (W Crow , USDA), the first drought indices really related to what was happening (A Al Bitar detecting the drought in California in 2011 when the official drought index was to detect it only a couple of years later) or monitoring the Mississippi  floods and levees destruction in 2011, the making of a flood risk forecasting tool demonstrator, the Spanish BEC fire risk analysis tool, etc… etc.. etc…

There isn’t enough room in a blog to document all this so I am giving only three samples.

1) high resolution soil moisture map

One of the main limitations of passive microwave is the spatial resolution. Olivier Merlin and his team developed an approach which -in many cases enables to monitor soil moisture with a 1 km resolution as shown in the example below.


Caption: 1 km soil moisture map from SMOS/ MODIS over Morocco (J. Malbeteau)

It can be successfully applied at 100 m in some cases (irrigation optimisation) as shown Catalonia (MJ Escorihuela). Other approaches rely on the use of active systems as originally planned for SMAP (N. Das) and done with SMOS (S. Tomer) or SMAP with Sentinel 1. Ideally the two approaches should be merged to my feeling.

Uses for such derived high resolution products are obvious, for irrigation and hydrology as already mentioned, but also for pest control (Locusts in Africa) or epidemiology (dengue, zika and malaria to name but a few). Moreover it can be used to derive high resolution root zone soil moisture and other passive L band products.

2) Rainfall estimates over land

It is known that rainfall mission (TRMM to GPM) are very useful tool for estimating rainfall distribution over land. It is also well known that estimating rainfall with one instantaneous measurement every so often is somewhat difficult. Sometimes and in some areas/context, the cumulated errors amount to several folds. The idea is thus to assimilate soil moisture estimates so as to « correct » the GPM rainfall estimates. Pellarin, Brocca and Crow and others demonstrated the efficiency of this approach.


Caption: Evolution of rainfall estimates after assimilating SMOS data (Pellarin, Brocca, Crow et al.)

3) Yield estimates

Soil moisture is a driven of crop yield in many areas. First shown by B. Hornebuckle with SMOS, Gibon and Pellarin went one step further by identifying which soil moisture (30 cm deep) and which period (grain filling and to a lesser extent reproductive) of vegetation growth where the drivers for millet in Western Africa. They then compared their local estimates with FAO global maps and found excellent correlation. It is interesting to see that departures are linked to local events


Caption: Soil moisture anomalies during two key stages and FAO Millet yield anomalies (F. Gibon)

Examples like this can be multiplied, I just picked some low hanging fruit. One can say that such applications an science results could be expected  and were delivered in record time. This blog is probably already way too long and I did not cover very interesting and promising results on evapotranspiration for instance, or hydrology, not to mention cryosphere … I keep the latter for tomorrow!

Stay tuned !

Further reading:

Brocca, L., Pellarin, T., Crow, W.T., Ciabatta, L., Massari, C., Ryu, D., Su, C.H., Rudiger, C., & Kerr, Y. (2016). Rainfall estimation by inverting SMOS soil moisture estimates: A comparison of different methods over Australia. Journal of Geophysical Research-Atmospheres, 121, 12062-12079

Molero, B., Merlin, O., Malbeteau, Y., Al Bitar, A., Cabot, F., Stefan, V., Kerr, Y., Bacon, S., Cosh, M.H., Bindlish, R., & Jackson, T.J. (2016). SMOS disaggregated soil moisture product at 1 km resolution: Processor overview and first validation results. Remote Sensing of Environment, 180, 361-376

Reul, N., Chapron, B., Zabolotskikh, E., Donlon, C., Quilfen, Y., Guimbard, S., & Piolle, J.F. (2016). A revised L-band radio-brightness sensitivity to extreme winds under tropical cyclones: The 5 year SMOS-Storm database. Remote Sensing of Environment, 180, 274-291

Roman-Cascon, C., Pellarin, T., Gibon, F., Brocca, L., Cosme, E., Crow, W., Fernandez-Prieto, D., Kerr, Y.H., & Massari, C. (2017). Correcting satellite-based precipitation products through SMOS soil moisture data assimilation in two land-surface models of different complexity: API and SURFEX. Remote Sensing of Environment, 200, 295-310.

SMOS catches rare events

Category : Data, L1

By Joaquin Munoz Sabater (ECMWF) and Fernando Martin Porqueras (IDEAS)

The 18 of April 2011 a strip with unusual values of brightness temperatures was detected in Western Sahara and Morocco. This feature looked strange enough to the Quality Check team Given at ESAC that they asked US what we thought about it. Many options were quickly discarded (RFI, Instrument problem etc…). The problem was that the measurements as shown on the QC picture below pointed towards high soil moisture which is not all that expected at this period of year in this area with such a linear shape.


Fig 1 QC picture on two orbits (Ascending and Descending) on April 18th 2011 from DPGS (X pol on the left and Y on the right)

However, between the 17 and 19 April 2011 a rare precipitation event took place in Western Sahara. Precipitations up to 10 mm, locally stronger, took place in a narrow, well defined band through the desert. This was caused by the influence of a low placed in the Azores which produced unusual precipitations in the Sahara area. This precipitation band moved progressively towards the North-East direction. Figures 1-4 show the cumulated total precipitation during 48 hours, in steps of 12 hours, starting the 16 April 2011. These figures clearly show the direction and cumulative values of the precipitation band. Non-coincidentally, there is a very good correlation between SMOS measured brightness temperatures (being much colder than its surroundings) and these precipitation figures, which explain these abnormal values. Figure 5 shows the ECMWF soil moisture analysis the 18 April 2011 at 12h00, for the first layer (7cm), which also displays this geographical strip being wetter (up to 20%) than its very dry surroundings. SMOS was able to catch very well this precipitation event.

step48 step49
step50 step51

These results demonstrates once more the good skills of the SMOS instrument to clearly catch rare precipitation events.

A first ever view of « below a hurricane » with SMOS

Category : CATDS, L3, Ocean

SMOS has opened a new field of research and application (one more!)

Observing sea surface state under a Hurricane

By Joe Tenerelli (CLS Brest)

It is well known that the L-band brightness temperatures measured by a downward looking radiometer such as that on board SMOS are significantly influenced by a number of radiation sources. Among the most important sources of L-band brightness over the ocean are:

1) flat surface emission (with order of magnitude 100 K);
2) Atmospheric emission (on the order of 5 K including reflected downwelling and upwelling);
3) scattered galactic radiation incident at the surface (order of magnitude 10 K);
4) excess emission associated with wind-driven surface roughness (order of magnitude 10 K).

By formulating a forward model of scene brightness for each of these sources, we can remove all but one of these contributions in order to reveal individual sources. For example, by removing all but the flat surface emission and by linearizing the salinity dependence about some reference value, we can retrieve deviations of surface salinity from that reference.
Similarly, by removing all but the scattered galactic noise contribution and projecting the remaining brightness onto the celestial sphere, we can obtain maps of scattered galactic noise as a function of surface wind speed (see an earlier post for results of this exercise).
Analogously, we can remove all but the rough surface emission to reveal the impact of roughness on the brightness temperatures at the radiometer. When we do this, we find a significant residual signal that continues to steadily increase up through hurricane force winds.
Below are two maps showing the first Stokes parameter (divided by two) of this residual roughness brightness. The residuals have been averaged over the dwell lines formed as the satellite passes over fixed points on earth. To the left we see this residual for Hurricane Earl at 2300 UTC 02 September, when it was a category 3 hurricane with maximum surface wind speed of 100 kt (about 51 m/s or 115 miles/hour). The residual first Stokes parameter (divided by two) reaches nearly 20 K to the right of the storm, which is a very significant signal and well above the per-dwell line noise level.
Similarly, the residual first Stokes (Tx+Ty)/2 for hurricane Igor near 0900 UTC 17 September is shown below and to the right, when it was a category 3 hurricane with maximum sustained surface wind speed of near 110 kt (about 55 m/s or 125 miles/hour). As with Earl, the roughness residual (Tx+Ty)/2 reaches nearly 20 K.


Since the SMOS interferometric radiometer, only samples a limited range of spatial frequencies, the spatial resolution of reconstructed brightness temperature maps is limited and features with scales less than about 50 km are strongly smoothed. Therefore, in order to compare SMOS roughness residual brightness temperatures to the HWIND wind fields we applied an average SMOS synthetic beam weighting function to the HWIND wind field in order to derive a smoothed wind field. Below we show various cross sections through the unsmoothed (left) and smoothed (right) HWIND analysis for 1930 UTC 16 September. The large impact of this smoothing is evident: Compared to the unsmooth wind field, the radius of maximum wind speed increases by a few kilometers, the maximum wind speed decreases from 100 kt to about 90 kt, and the minimum surface wind speed at the storm center increases to about 50 kt.
Strictly speaking, the problem of relating synthetic beam smoothed brightness temperatures to ‘smoothed’ wind fields is ill-posed, since, without any assumptions, one can generally find multiple distributions of unsmoothed wind fields that yield different smoothed brightness temperature fields for the same smoothed wind field. However, to the extent that the roughness residual brightness temperature varies linearly with wind speed, we can justifiably derive a ‘smoothed’ model by relating a smoothed wind field to synthetic beam averaged brightness temperature field.


Examining more closely the roughness residual first Stokes parameter for the hurricane Igor overpass on 17 September, we see that the pattern of L-band brightness residual closely resembles that of the near-surface wind field derived by the NOAA AOML Hurricane Research Division (HRD) based in Miami, Florida. In the figure below and to the left, we show the residual first Stokes parameter from SMOS in a cartesian coordinate system with origin at the storm center. To the right we show the HRD surface wind field analysis, derived using their HWIND analysis system, for 1930 UTC 16 September. In this case (but not always) the HWIND analysis field incorporates (among many other sources of data) surface wind speeds derived (along with rain rate) from a C-band Stepped Frequency Microwave Radiometer (SFMR). For excellent descriptions of the wind and rain retrieval algorithms for the SFMR see the papers of Uhlhorn et al. and Jiang et al., Monthly Weather Review 2003, 2006, and 2007. Other sources of wind information for this analysis include GPS sondes, an ASCAT pass, and a  buoy time series.

To derive an empirical relationship between residual SMOS total power and surface wind speed we use two HWIND surface wind fields derived for 1930 UTC on September 16 and 17. Unfortunately, the analyses are separated by a day with the SMOS overpass occurring midway between them, and in this time period the storm was beginning to weaken. To arrive at a very rough approximate wind field for deriving a model we take the average of the two HWIND analyses and we recognize that some error may be introduced in this averaging.

In deriving the model, we will assume that all of the residual brightness is related to roughness and that none is related to precipitation. At L-band, emission and attenuation from precipitation if much lower than at higher frequencies; nevertheless, it is clear that the brightness temperatures may increase by several kelvin in the vicinity of rainband, and research involving collocated passive and active data at higher frequencies is required in order to address this issue.


The approach we take here is to simply find the function that maps the SMOS brightness temperatures into a ‘wind field’ with a probability distribution that matches that of the temporally averaged and spatially smoothed wind field obtained from the two HWIND analyses. This matching considers only winds within 300 km of storm center and the domain is shown in the third figure above.

Below and to the left we show the CDFs obtained from the averaged HWIND analysis and those obtained from two SMOS empirical models. One using the exact PDF match and the other using a linear fit to the model derived from the PDF match. Also shown are the minimum wind speeds for tropical storms and hurricanes.

Below and to the right we show the roughness residual first Stokes parameter (divided by two) as a function of wind speed for both the exact (dashed black) and linear (solid black) PDF matching models. We also show linear extrapolations of an empirical model derived from SMOS data for wind speeds below about 23 m/s. As the hurricane model is based on dwell line averaged brightness, no direct comparisn can be made between the hurricane model and the lower wind speed model, but  there is significant discrepancy between the models for all incidence angles which evidence of the special nature of sea states within hurricanes.


Below we compare the wind field derived using the hurricane L-band wind speed model (left) to the SMOS-smoothed wind field from the time-averaged averaged HWIND analysis (right).

By construction, the SMOS wind field captures the maximum wind speed, but the spatial structures in the two wind fields are somewhat different. First of all, the asymmetry across track is larger in the SMOS wind field than in the HWIND analysis, and this is probably related to variability in sea states for a given surface wind speed.

Secondly, a band of high wind speed to the SW of storm center is evident in the SMOS wind field, and this is likely associated with a rainband that is evident in higher frequency brightness temperature maps from an AMSR overpass.

Despite these discrepancies, it is clear that SMOS can capture the essential features of strong tropical cyclones, but to obtain a useful product more work is required to understand the relations between wind speed and sea state and the associated impact on L-band brightness temperature.


Soil Moisture Temporal Evolution As Seen By SMOS

Category : CATDS, Commissioning phase, L2

January 8 to June 30 2010
embedded by Embedded Video

double click to enlarge

This small animation was produced in the framework of the Centre Aval de Traitement des Données SMOS (CATDS funded by CNES) by Arnaud Quesney (Capgemini) and Elsa Jacquette (CESBIO) from DPGS data (ascending orbits) using the L2SM retrieval algorithm and L3 CATDS aggregation.

The maps were produced with a 3 day running window and using a 3X3 median filter. It covers the period January 8 to June 30 2010 and is thus made with acquisitions covering all the stages / improvements encountered during the commissioning phase as the data has not been reprocessed yet. This can be inferred by just looking at the temporal evolution!
One may not the RFI impact in some areas as well as the substantial decrease of RFI impact in Western Europe as the sources in Spain were eradicated.
Also have a look at Australia. You might see the rain system sweeping across the continent by th ened of th eperiod. You may also see the melt progressing at high latitudes…., monsoon effects etc….

A word of warning though. this film is for wetting your appetite only. The data was obtained with varying configurations (full and dual pol for instance), with several versions of the calibration scheme, of the processors (both level 1 and Level 2), so we need to have the data reprocessed before we can analyse in depth such information.
The thing to remember is that it can only improve and it is already quite impressive…..

download wordpress themes