Moisture in the topsoil: From large-scale observations to small-scale process understanding

(1)

(2)

(3)

Moisture in the topsoil

From large-scale observations to small-scale

process understanding

ter verkrijging van de graad van doctor aan de Technische Universiteit Delft,

op gezag van de Rector Magnificus prof. ir. K.C.A.M. Luyben, voorzitter van het College voor Promoties,

in het openbaar te verdedigen op 5 maart 2015 om 10:00 uur

door

Martine Marije RUTTEN

Master of Science

(4)

This dissertation has been approved by the

Promotor: Prof. dr. ir. N.C. van de Giesen Delft University of Technology Composition of the doctoral committee:

Rector Magnificus Chairperson

Dr. S.C. Steele-Dunne Delft University of Technology Independent members:

Prof dr. J. Selker Oregon State University Prof. dr. ir. W.S.J. Uijttewaal Delft University of Technology Prof. dr. W.G.M. Bastiaanssen UNESCO-IHE and Delft University of

Technology

Prof. dr. ir. T.J. Heimovaara Delft University of Technology Dr. W.P. Breugem Delft University of Technology

(5)

Summary

The boundary that separates the earth from the atmosphere is a crucial zone of study for

meteorology and hydrology. Here, solar energy is partitioned into sensible heat which drives atmospheric circulation, latent heat needed for evaporation from the soil and transpiration of vegetation, and soil heat which warms the subsurface. Precipitation is partitioned into

interception that evaporates directly into the atmosphere, surface runoff that discharges quickly into water courses and infiltration which resides longer in the subsurface. Soil moisture

influences all these processes and is therefore considered a key variable in land-atmosphere interaction.

In order to obtain a better understanding of the heat and water balance of topsoil, observations are key, but challenging with in situ point sensors. Recent rapid developments in remote sensing have tremendously increased our ability to observe the boundary between soil and atmosphere. Retrieving state variables such as soil temperature and moisture from remote sensing is far from trivial: detected signals originate not only from the soil, but also from the atmosphere and vegetation, the depth of the detection is a function of the soil moisture itself, and pixels are large and heterogeneous. Field validation is difficult, because of scale disparity between in situ point sensors and remote sensing pixels. Still, given the limitations, remote sensing provides an opportunity to improve understanding of heat and moisture transfer in the topsoil. The central question of this research is:

What can be learnt from (remote sensing) observations about the heat and moisture balance of the topsoil?

First a cross validation of different soil moisture products based on remote sensing was performed to investigate similarities and differences between these products. The differences were significant and could be attributed to differences in land use and vegetation, but not fully explained. This illustrated that retrieval algorithms for soil moisture are far from converged. One prerequisite for improving retrieval algorithms is ground truth, ground observations at scales relevant for remote sensing.

Second, a field technique was developed that can potentially be used for bridging the observation gap between point sensors and remote sensing pixels. This technique uses Distributed Temperature Sensing (DTS) over horizontal extents up to kilometers to infer soil moisture at this intermediate scale. Propagation of variations in atmospheric temperature and radiation with depth is a function of soil moisture. By using DTS observations at three depths, it is possible to infer soil moisture, assuming that heat conduction is the dominant heat transfer mechanism. The heat diffusion equation is inverted to obtain estimates of soil heat diffusivity and soil moisture. Since this technique relies on observations of the passive thermal response of the soil to atmospheric temperature and radiation variations, this technique is called passive SoilDTS in contrast to active soil DTS, which relies on active heat pulses. The feasibility of passive SoilDTS for soil moisture estimation was asserted in a field experiment conducted in Monster in the Netherlands.

The analysis of the experimental results of this feasibility study pointed out a number of technical and modeling issues that needed to be investigated further in order for passive SoilDTS to be used for soil moisture estimation and scaling. Some soil moisture estimates were not reasonable due to uncertainties in cable depths and heat transfer mechanisms. To separate the technical issue of cable depth from the modeling challenges, the same methodology used to infer soil moisture from passive SoilDTS was applied to profile data of temperature and soil moisture obtained with point sensors. The depth of these point sensors could be determined with far greater accuracy than the cable depth. Analysis of the point observations challenged the common assumption that conduction is the dominant heat transfer mechanism in soil. Evaporation seemed to play a dominant role in heat transfer on dry days. Yet evaporation rates

(6)

found were higher than would be expected if mass diffusion would be the dominant transfer mechanism of water vapor. Vapor diffusion appeared to be enhanced.

Enhancement of vapor diffusion is a long-studied phenomenon, subject to debate on the explanations of underlying mechanism. In an extensive literature review on vapor enhancement in soils, the plausibility of various mechanisms was assessed. We reviewed mechanisms based on (combinations of) diffusive, viscous, buoyant, capillary and external pressure forces including: thermodiffusion, dispersion, Stefan’s flow, Knudsen diffusion, liquid island effect, hydraulic lift, free convection, double diffusive convection and forced convection. The analysis of the order of magnitude of the mechanisms based on first principles clearly distinguishes between plausible and implausible mechanisms. Thermodiffusion, Stefan’s flow, Knudsen effects, liquid islands do not significantly contribute to enhanced evaporation. Double diffusive convection seemed unlikely due to lack of experimental evidence, but could not be completely excluded from the list of potential mechanisms. Hydraulic lift, the mechanism that small capillaries lift liquid water to the surface where it evaporates, does significantly contribute to enhanced evaporation from soils, also from dryer soils. The experimental evidence for and the theoretical underpinnings of this mechanism are convincing. However, we sought

mechanisms that both explain enhanced evaporation and steep temperature gradients in the soil during the daytime. These often observed gradients consist of a sharp decrease of

temperature with a depth up to the depth of the evaporation front. Hydraulic lift cannot explain this because the evaporation front is located at the surface. One remaining mechanism is forced convection due to atmospheric pressure fluctuations, also referred to as wind pumping. Wind pumping causes displacement and flow velocities too small for significant convective and too small for significant dispersive transport, when steady state dispersion formulations are used. However, experiments do indicate significant dispersive transport that can be explained by dispersion under unsteady flow conditions. Forced convection due to pressure fluctuations seems to be the only mechanism that can explain both enhanced evaporation and the steep temperature gradients.

We investigated under which conditions wind pumping can enhance water vapor transfer from the soil to the atmosphere and which mechanisms are responsible for this enhancement in a modeling study. Previous models of wind pumping relied on enhanced transfer due to enhanced mixing described with empirical macroscopic dispersion coefficients with weak physical

foundations. We searched for better understanding of physical mechanisms driving enhanced mixing. With combination of order of magnitude analysis, phenomenological, empirical and analytical models, mechanisms were investigated. A model for surface pressure fluctuations was coupled with a pressure diffusion model, a pore flow velocity model and a dispersion model. Based on this coupled model, we propose that the enhancement is caused by mixing at the pore level due to flow instabilities. Fast pressure fluctuations at the soil-air interface make vortices in the soil unstable. Instabilities arise when the timescale of the pressure fluctuations is close to the timescale of viscous dissipation which is related to the pore size. In this case, vortices in the soil cannot increase, decrease and turn direction, in phase with the pressure fluctuations and instabilities occur in the form of ejections. The ejections of vortices enhance mixing and transport. Timescales of wind induced pressure fluctuations and pore sizes are such that this mechanism is considered likely in soils. Further research is needed to prove this mechanism and quantify it. The developed model is a hypothesis and should be tested with numerical and laboratory experiments. For estimating the effect of this vapor enhancement on the soil heat budget, a coupled heat and moisture transfer model should be developed. Such a model could also shed light on the relative importance of hydraulic lift and wind pumping for evaporation rates.

Perhaps, because the topsoil forms the boundary between land and atmosphere, but also between two disciplines meteorology and hydrology, there are still many questions that remain about heat and moisture transfer in the upper few centimeters of the soil. Remote sensing soil

(7)

a result, remote sensing presents not only a challenge for ground validation, but also an opportunity for hydrological and meteorological model improvement. Observation is the beginning of most learning.

(8)

(9)

Samenvatting

De grens die de aarde scheidt van de atmosfeer is een cruciaal gebied van studie voor de

meteorologie en hydrologie. Hier wordt de zonne-energie verdeeld in voelbare warmte, die de atmosferische circulatie drijft, latente warmte, die nodig is voor de verdamping uit de bodem en transpiratie van de vegetatie en bodemwarmte, die de ondergrond opwarmt. Neerslag wordt verdeeld in interceptie, die direct verdampt in de atmosfeer, oppervlakkige afstroming, die snel loost in waterlopen, en infiltratie, die zich langer in de ondergrond bevindt. Bodemvocht beïnvloedt al deze processen en wordt daarom beschouwd als een belangrijke variabele in de wisselwerking tussen land en atmosfeer.

Om een beter begrip van de warmte en waterbalans van het aardoppervlak te verkrijgen zijn waarnemingen essentieel, maar uitdagend om te verkrijgen op grote schaal met in-situ

instrumenten. Recente snelle ontwikkelingen in de aardobservatie vanuit de ruimte hebben het vermogen om de grens tussen de grond en de atmosfeer te observeren significant vergroot. Het observeren van bodemtemperatuur en bodemvocht met aardobservatie is verre van triviaal: gedetecteerde signalen zijn niet alleen van de bodem zelf afkomstig, maar ook van de

tussenliggende atmosfeer en de vegetatie; de diepte van de waarneming is afhankelijk van de bodemvochtigheid zelf; en pixels zijn groot en heterogeen. Veldvalidatie is moeilijk vanwege het schaalverschil tussen in situ instrumenten, die de toestand op één punt meten, en satelliet instrumenten, die de toestand gemiddeld over een pixel meten. Echter, nog steeds, ook gezien de beperkingen, biedt aardobservatie een kans om het begrip van de warmte- en

vochthuishouding in de bovengrond te verbeteren. De centrale vraag van dit onderzoek luidt: Wat kan worden geleerd van observaties van onder andere instrumenten op satellieten over de warmte- en de waterbalans van de bovengrond?

Als eerste is kruisvalidatie van verschillende bodemvochtproducten op basis van

satellietinstrumenten uitgevoerd om overeenkomsten en verschillen te onderzoeken. De verschillen waren significant en konden deels worden toegeschreven aan verschillen in

landgebruik en vegetatie, maar niet volledig worden verklaard. Dit illustreert dat algoritmes, die gebruikt worden om bodemvocht te bepalen op basis van de satellietwaarneming, verre van geconvergeerd zijn. Een voorwaarde voor het verbeteren van deze algoritmes is het verkrijgen van grondwaarnemingen op schalen die relevant zijn voor aardobservatie met instrumenten op satellieten.

Een veldtechniek is ontwikkeld die mogelijk kan worden gebruikt voor het overbruggen van de kloof tussen puntsensoren en aardobservatie pixels. Deze techniek maakt gebruik van

Distributed Temperature Sensing (DTS) over grote horizontale afstanden om bodemvocht af te leiden bij deze intermediaire schaal. De voortplanting van variaties in atmosferische

temperatuur in de ondergrond is een functie van de bodemvochtigheid. Uit DTS waarnemingen op drie dieptes kan bodemvocht worden afgeleid onder de aanname dat geleiding het

dominante mechanisme is voor warmteoverdracht. De diffusievergelijking voor temperatuur wordt geïnverteerd om schattingen van de warmtediffusiviteit van de bodem en vervolgens bodemvocht te verkrijgen. Omdat deze techniek gebaseerd is op waarnemingen van de reactie van de bodem op atmosferische temperatuur en straling, wordt deze techniek passieve SoilDTS genoemd in tegenstelling tot actieve SoilDTS, wat gebruikt maakt van actieve warmte pulsen. De haalbaarheid van passieve SoilDTS voor bodemvochtschatting is getest in een

veldexperiment uitgevoerd in Monster in Nederland.

De analyse van de experimentele resultaten van deze haalbaarheidsstudie wees op een aantal technische en modellering aspecten die verder worden onderzocht moeten worden om passieve SoilDTS te gebruiken voor bodemvochtigheidschatting. Sommige bodemvochtschattingen waren niet redelijk, te wijten aan onzekerheden in de kabeldieptes en warmteoverdrachtmechanismen.

(10)

Om het technische probleem van kabeldiepte te scheiden van de modelleeruitdagingen, werd dezelfde methode gebruikt voor het afleiden van bodemvochtigheid uit passieve SoilDTS

toegepast op de profielgegevens van temperatuur en bodemvocht verkregen met puntsensoren. De diepte van deze puntsensoren kon bepaald worden met een veel grotere nauwkeurigheid dan de diepte van de kabels in de veldstudie in Monster. Analyse van de meetresultaten betwiste de gebruikelijke aanname dat geleiding het dominante warmteoverdrachtmechanisme is in de bodem. Verdamping lijkt een dominante rol in de warmteoverdracht te spelen te spelen, ook op droge dagen. De verdamping was echter groter dan zou worden verwacht wanneer massadiffusie het dominante transportmechanisme voor waterdamp zou zijn. Er lijkt sprake van vergroot damptransport.

Vergroot damptransport is fenomeen dat al decennia bestudeerd wordt en er is discussie over de onderliggende mechanismen. In een uitgebreide literatuurstudie naar vergroot

damptransport in de bodem, is de plausibiliteit van verschillende mechanismen beoordeeld. De beoordeelde mechanismen zijn (combinaties van) diffusie, viskeuze, capillaire en externe krachten waaronder: thermodiffusie, dispersie, Stefan's flow, Knudsen diffusie, vloeibare-eiland effect, hydraulische lift, convectie, dubbele diffusieve convectie en gedwongen convectie. Een analyse van de orde van grootte van de mechanismen maakte duidelijk onderscheid tussen aannemelijke en minder aannemelijke mechanismen. Thermodiffusion, Stefan's flow, Knudsen diffusie en vloeibare eilanden kunnen niet significant bijdragen tot een verhoogde verdamping. Dubbele diffusieve convectie lijkt onwaarschijnlijk wegens gebrek aan experimenteel bewijs, maar kon niet volledig worden uitgesloten. Hydraulische lift, een mechanisme waarbij kleine capillairen vloeibaar water aan het oppervlak tillen waar het verdampt, kan in belangrijke mate bijdragen tot een verhoogde verdamping uit de bodem, ook uit droger bodems. Het

experimentele bewijs voor en de theoretische onderbouwing van dit mechanisme zijn overtuigend. Echter, de zoektocht in dit onderzoek richt zich op mechanismen die zowel vergroot damptransport als steile temperatuurgradiënten in de bodem verklaren. Deze vaak waargenomen gradiënten bestaan uit een scherpe daling van de temperatuur met diepte tot het verdampingsfront. Hydraulische lift kan dit niet verklaren, omdat de verdamping aan de

oppervlakte plaatsvindt. Een resterend mechanisme is gedwongen convectie door atmosferische drukschommelingen, ook wel windpompen genoemd. Windpompen veroorzaakt verplaatsing en stroomsnelheden te klein voor significante convectie en te klein voor significant dispersief transport, wanneer stationaire dispersieformuleringen worden gebruikt. Echter, experimenten geven aan dat significant dispersief transport kan worden verklaard door dispersie onder onstabiele stromingscondities. Windpompen lijkt het enige mechanisme dat zowel vergroot damptransport als steile temperatuurgradiënten kan verklaren.

In een modelstudie is verder onderzocht onder welke voorwaarden windpompen

waterdamptransport van de bodem naar de atmosfeer kan vergroten, welke mechanismen hiervoor verantwoordelijk zijn. Bestaande modellen van windpompen verklaren vergroot

damptransport op basis van sterk empirische macroscopische dispersiecoëfficiënten met zwakke fysische onderbouwing. We zochten beter begrip van de fysische mechanismen achter de verhoogde menging. Met een combinatie van de grootteorde analyse, fenomenologisch, empirische en analytische modellen, werden mechanismen onderzocht. Een model voor windpompen op basis van een atmosferisch turbulentie model, een luchtstromingsmodel voor de grond en een dispersiemodel is ontwikkeld. Op basis van dit gekoppelde model, stellen we de hypothese dat vergroot damptransport wordt veroorzaakt door versterkte menging op porieniveau door stromingsinstabiliteit. Snelle drukfluctuaties maken stromingspatronen in de bodem instabiel als de tijdschaal van de fluctuaties dezelfde ordegrootte heeft als de tijdschaal van viskeuze demping van deze fluctuaties. In dit geval kunnen wervels in de bodem niet vergroten, verkleinen en van richting veranderen in fase met de drukfluctuaties en instabiliteit optreden die menging veroorzaakt. Tijdschalen van drukfluctuaties en poriën in de bodem zijn zodanig dat dit mechanisme waarschijnlijk geacht wordt in de bodem. Verder onderzoek is nodig om dit mechanisme te bewijzen en te kwantificeren is. Het ontwikkelde model is een

(11)

schatten van het effect van dit vergroot damptransport op de bodemtemperatuur, dient een gekoppeld water en warmtetransport model ontwikkeld te worden. Een dergelijk model zou ook licht werpen op het relatieve belang van de hydraulische lift en windpompen voor verdamping. Misschien, omdat de bovengrond de grens vormt tussen land en atmosfeer, maar ook tussen twee disciplines meteorologie en hydrologie, zijn er nog vele vragen die blijven bestaan over de warmte en vochttransport in de bovenste paar centimeter van de grond. Aardobservaties van bodemvocht met satellietinstrumenten dwingen de wetenschappelijke gemeenschap om begrip van de bovengrond te herzien. Aardobservatie biedt niet alleen een uitdaging voor de

grondvalidatie, maar ook een kans voor hydrologische en meteorologische modelverbetering. Observatie is het begin van het meeste leren.

(12)

(13)

6.1. Introduction……….101 6.2. Conceptual understanding………103 6.3. Mathematical model………...107 6.4 Results……….109 6.5. Discussion………113 6.6. Conclusion………114 6 FINAL DISCUSSION………..…..115 REFERENCES…………..………..…..119 CURRICULUM VITAE………131

(14)

(15)

Nomenclature

a constant

A amplitude

b Klinkenberg factor

c specific heat capacity [J kg-1 _K-1_{], molecular heat capacity [J K}-1_{] relative concentration}

C volumetric heat capacity [Jm-3_K-1_{], empirical constants, concentration [mol m}-3_]

d pore size [m] D diffusivity [m2_s-1_]

e vapor pressure [Pa or mBar]

E emissivity, or evaporation flux [m s-1_]

f frequency [Hz], factor Fo Fourier number

g gravity acceleration [ms-2_]

G soil heat flux [W m-2_]

h relative humidity H sensible heat [W m-2_]

j specific flux [mol m-2_s-1_]

k permeability [m2_]

kB Boltzman contant

kT ratio of mass diffusion coefficient and thermal diffusion coefficient [-]

K air bulk modulus

l length scale [m], mean free path length [m] L latent heat of vaporization [Jkg-1_{], soil depth [m]}

LE latent heat [W m-2_]

m molecular mass [kg] M molar mass [kg mol-1_]

n porosity, molecule concentration [m-3_]

p pressure [Pa] P period [s] Pe Peclet number Q quartz content, Flux

r tube radius [m], resistance [sm-1_]

R radiation [W m-2_{] or universal gas constant [Jmol}-1_K-1_]

RT ratio of inner to outer boundary layer

Ra Rayleigh number Re Reynolds number RH relative humidity S sink term

SM satellite retrieved soil moisture T temperature [K or C] U wind speed [m s-1_] *

u

friction velocity [m s-1] v flow velocity [m s-1_] z depth [m]

(16)

Greek



pressure diffusivity [m2_s-1_{] or dispersion length [m], thermal diffusion constatn}



thermal expansion coefficient [K-1_]

C



concentration expansion coefficient [mol-1_]



adiabatic index , surface tension [N m-1_]



Stokes layer thickness [m]



energy dissipation rate [m2_s-3_]



characteristic depth of diffusion [m]



Kolmogorov micro scale [m]



soil moisture content, contact angle



thermal conductivity [Wm-1_K-1_]



dynamic viscosity [Pas], chemical potential [Jmol-1_]



kinematic viscosity [m2_s-1_]



scaling factor



density [kg m-3_]



Stefan-Boltzman constant [kg s-3 _K-4_]



tortuosity, timescale [s]



power spectral density



angular frequency [rad s-1_]

Subscript 0 surface

a air, atmosphere, airodynamic c critical, roll over (frequency) down downwelling (radiation) enh enhancement h heat K Knudsen l lower lam laminar L longwave m bulk opt optimal u upper up upwelling (radiation) tur turbulent v vapor w water

(17)

1

Introduction

1.1. The significance of the topsoil and soil moisture

The boundary that separates the earth from the atmosphere is a crucial zone of study for meteorology and hydrology. Here, solar energy is partitioned into sensible heat which drives atmospheric circulation, latent heat needed for evaporation from the soil and transpiration of vegetation, and soil heat which warms the subsurface. Precipitation is partitioned into

interception that evaporates directly into the atmosphere, surface runoff that discharges quickly into water courses and infiltration which resides longer in the subsurface. The processes in Figure 1.1 are influenced by soil moisture. Therefore, soil moisture has been identified as a key state variable in hydrology and in land-atmosphere interaction which is important to observe at various spatial and temporal scales. Climate models and numerical weather prediction models need relatively coarse scale observations, while fine scale observations are required for more detailed (agro) hydrological models.

(18)

1.2. Observing the topsoil

In order to obtain a better understanding of the heat and water balance of topsoil, observations are key, yet challenging with in situ point sensors. Heat flux plates and hydrological sensors need to be installed at a sufficient depth in order for the surface roughness or sensor

dimensions not to disturb the measurements. In addition, ground observation networks are not extensive, covering only a few square meters of the earth’s surface.

Observing soil moisture in situ is a challenge of its own. Gravimetric soil moisture sampling gives a good indication of the true soil moisture quantities, but it is difficult to obtain a time series with high temporal resolution using this technique. Therefore, often the proxies heat and electrical conductivity and/or capacity are used (e.g. Campbell et al., 1991; Dalton et al., 1984). The wetter the soil, the faster a heat pulse travels through the soil, because the soil’s thermal conductivity is positively related to its soil moisture content. Moreover, the wetter the soil, the faster the heat pulse is reduced since the soil’s thermal capacity is negatively related to soil moisture. Similarly, the wetter the soil, the faster an electricity pulse travels in the soil and the faster the electricity pulse is reduced. Heat and electrical proxies of soil moisture are not only used in in situ sensors on the ground, but also form the basis for remote sensing retrievals of soil moisture.

Recent rapid developments in remote sensing measurements have tremendously increased our ability to observe the boundary between the ground and the atmosphere. The earth emits radiation at a range of wavelengths. Depending on the wavelength, the emitted radiation is sensitive to various properties in the atmosphere, surface and subsurface as illustrated in Figure 1.2. Visible and shorter infrared wavelengths are the result of solar radiation instantaneously emitted or reflected when it hits a surface and are useful for applications such as cloud

detection and land cover. Longer infrared wavelengths are the result of adsorbed solar radiation emitted as heat and are useful for temperature observations of the surface. Microwaves are emitted in a similar way as thermal infrared radiation and are useful for moisture observations of the topsoil.

The longer the wavelength, the deeper the zones are from which microwaves are emitted. C-band retrievals can be used to estimate moisture in the top few centimeters of the soil, whereas L-band retrievals can be used to estimate soil moisture in the top ten centimeters (Entekhabi et al., 2010). One clear advantage of microwaves is that, due to the larger wavelength, they can penetrate clouds and the largest microwaves (L-Band) can even penetrate vegetation. Yet one disadvantage of microwaves is that the total energy emitted is very small. Hence passive microwave detection is only possible with a low spatial resolution in the order of 50 km. Active remote sensing avoids this issue of limited emitting energy. With active microwave remote sensing, microwave signals are emitted from the satellite platform to the earth’s surface and sensors detect the reflection. This technique has proven to be very useful for soil moisture detection. However, the increase in spatial resolution that can be achieved with active sensors comes at the expense of temporal resolution. Since with greater spatial resolution, the narrower the swat and the larger the revisit time becomes which can result in lower temporal resolution.

(19)

Figure 1.2: Sketch of the electromagnetic spectrum and its use for satellite remote sensing. The passive (active) sensors relevant for this thesis are indicated in boxes with thin (thick) lines. These include the Visible and Infrared Scanner (VIRS), Precipitation Radar (PR) and TRMM Microwave Imager (TMI) onboard the Tropical Rainfall Measuring Mission (TRMM), the European Remote Sensing (ERS) satellite, Advanced Microwave Scanning Radiometer(AMSRE), Soil Moisture and Ocean Salinity Mission (SMOS), Satellite Pour l'Observation de la Terre (SPOT) and Landsat.

At the molecular level, the working principle of microwave soil moisture remote sensing relies on the electrical properties of water. Water is a strong dielectric, a good store for microwave energy. The relative permittivity (also called dielectric constant, ratio of material permittivity to vacuum permittivity) of water is in the order of 80, whereas the relative permittivity of dry soil is lower than five. As such, the soil relative permittivity is a strong function of soil moisture. The relative permittivity governs the emissivity and reflectivity of the soil. Passive soil moisture remote sensing uses the reduction in emissivity, because the relative permittivity of the soil increases as the moisture level in the soil increases. In contrast, active remote sensing uses the reflectivity of the soil, since the reflection of radar waves increases as the moisture level in the soil increases. Emissivity and reflectivity are related though Kirchhoff’s law of radiation which states that the sum of emissivity and reflectivity equals one, see also Figure 1.2.

(20)

In 2009, the European Space Agency (ESA) launched the Soil Moisture and Ocean Salinity (SMOS) mission (Kerr et al., 2001). This was the first mission dedicated to soil moisture observation. For a long time, soil moisture has been retrieved as a by-product from various sensors, including: the active C-band scatterometers on board of ERS-1, ERS-2 and METOP-A (Wagner et al., 1999), the active L-band radar on board the Advanced Land Observing Satellite (ALOS) (Shimada et al., 2010) and the passive C-band radiometers on board of Aqua satellite (Njoku et al., 2003; De Jeu et al., 2003). In 2014 or 2015, the National Aeronatics and Space Administration (NASA) plans to launch the Soil Moisture Active Passive (SMAP) mission (Entekhabi, 2010). SMAP consist of an active L-band radar and a passive L-band radiometer. Expectations are high for soil moisture retrieval from SMAP because L-band is the most suitable band for soil moisture detection and because the combination of active and passive sensors enables both high spatial and temporal resolution.

1.3. Retrieval algorithms and validation

Retrieving soil moisture from microwave observations whether passive or active is far from trivial. Detected signals originate not only from the soil, but also from the atmosphere and vegetation; the depth of the detection is a function of the soil moisture itself; and pixels are large and heterogeneous. Passive retrievals are especially sensitive to soil temperature which is also a function of soil moisture as illustrated in Figure 1.3. There are various approaches to soil moisture retrieval ranging from physically based models to statistical models.

Figure 1.3: Sketch of passive and active soil moisture retrieval showing how passive brightness temperature observations are influenced by ground effective

temperature and emissivity. Emissivity is a strong function of soil moisture and so is the soil temperature. Backscatter from active sensors is influenced by roughness

(21)

For example, for SMOS retrievals an iterative approach was chosen using a physical based forward model of the radiative transfer equations. Soil moisture and vegetation opacity (transparency of vegetation for microwaves) are adjusted until the brightness temperature (measure of the emissivity of microwave radiation), predicted with the forward model, matches the observed brightness temperature (Kerr et al., 2012). Critical is the estimation of the

effective temperature of the soil that together with the emissivity determines the brightness temperature as illustrated in Figure 1.3. For shorter wavelengths (X- and C-band) and thus shallow detection depths (1-2 mm), an onboard K-band sensor can be used to estimate ground temperature. SMOS does not carry an onboard K-band sensor and moreover K-band ground temperatures are not representative for the deeper detection depths (up to 20cm) that can be achieved with the L-band. Therefore, the SMOS algorithm relies on ground temperature products from the numerical weather prediction (NWP) system of the European Centre for Medium-Range Weather Forecasts (ECMWF) of the Tiled ECMWF Scheme for Surface Exchange over Land (TESSEL)(Holmes et al., 2012) This can be considered a serious weakness of the retrieval algorithm, because temperature profiles are strongly influenced by soil moisture content as discussed in Section 1.2. Moreover, soil temperatures also influence moisture and permittivity profiles, but this influence is only minor, therefore it is indicated with thin arrows in Figure 1.3.

For ERS retrievals, a statistical approach was chosen. The backscatter is a function of the reflectivity and roughness of the soil as illustrated in Figure 1.3. ERS retrieval algorithm is based on the variation in the reflectivity of the soil depending on soil moisture over time, while the roughness of the soil remains static over time. Hence, after the removal of the vegetation effects, the minimum backscatter observed over a certain period is assumed to be equal to dry soil, while the maximum backscatter is assumed to be equal to saturated soil and the soil moisture content for intermediate values is determined by linear interpolation (Wagner et al., 1999).

1.4. Assimilation and validation

Satellite-based temperature, soil moisture, and land cover retrievals can provide vast amounts of information at an unprecedented scale. Satellite retrievals of hydro-meteorological variables are increasingly being assimilated in models as a model state variable in addition to ground observations. Based on the premise that “the truth” lies between all available models and observations, the more models and observations available the better the predictions should become. Data assimilation schemes, such as the Kalman filter, weigh the error of models and observations to arrive at optimal predictions. For example, Hain et al. (2012) and Sahoo et al. (2013) show how the assimilation of satellite soil moisture retrievals can improve the soil moisture predictions of the Noah land surface model. Data assimilation can not only improve model predictions, it can also provide insight to systematic errors with either the observations or the models. Systematic errors indicate a deficiency in the retrieval algorithm and/or in the hydro-meteorological model. These deficiencies can then be reduced, often with the help of additional field or model experiments. This is a cycle of continual improvement.

The relations between the observed quantity and the modeled quantity are not straightforward. Often the spatial scale of models is smaller than the spatial scale of satellite observations and as a result the downscaling of these observations is required. Soil moisture can vary

significantly across the satellite footprint. For instance, point measurement within a 502 km2

scale, which represents a typical footprint scale, may present a standard deviation 0.06 ~ 0.08 m3_m-3_(Crow_{et al.}_{, 2012). Downscaling techniques are needed to take this spatial variability}

into account. In order to develop such downscaling techniques, extensive networks of soil moisture observations across scales are required. Differences in land cover across the pixel make downscaling even more challenging.

(22)

Moreover, observations of a very shallow topsoil layer need to be related to deeper soil layers such as the root zone. Questions arise such as: how does the soil moisture estimated in the top few centimeters of the soil relate to the root zone moisture (Crow et al., 2008; Das and

Mohanty, 2006)? How does skin temperature relate to soil temperature (Holmes et al., 2008)? The hydrological models (e.g. Hydrus) and Land Surface Models (e.g. TESSEL or Noah), commonly used in assimilation, were not designed to resolve the water and heat balance at the millimeter to centimeter scale. These models are macroscopic, treating soil matrix as a

continuum. The implicit assumption that underlies this approach is that the soil parameters over the smallest scale in the discretization, the Representative Elementary Volume (REV), can be described by a single value. This is questionable when the spatial scale of interest approaches the scale of pores, grains and cracks. Clausnitzer and Hopmans (1999) found that for mono-size glass beads the REV scale is approximately five times the bead diameter, while Al-Raoush et al. (2003) have shown that for the theoretical packing of multiple-size glass beads the REV scale is approximately 25 times the median diameter. For homogeneous well sorted soils the REV scale may be in the order of a couple of millimeters (Costanza-Robinson et al., 2011), but for natural soils this scale is more likely to be larger than one centimeter. For cracked soils, the REV scale is even larger and mostly determined by the crack length, this scale may even be close to five times the mean crack length (Li and Zgang, 2010).

1.5. Objectives and outline

It becomes clear from the discussion in the previous section that there are many challenges for using remote sensing observations to their full potential. Some of these challenges will be addressed in this thesis.

The first challenge is to make optimal use of the largely complementary characteristics of active and passive remote sensing and the different wavelengths. In the Chapter 2, soil moisture retrievals from active and passive remote sensing observations are compared and validated using remotely sensed precipitation over the Iberian Peninsula. The objective is to obtain more insight into the similarities and differences between active and passive sensors and between retrieval algorithms.

The second challenge is the scaling of soil moisture. Chapter 3 focuses on a field technique that can potentially be used for bridging the observation gap between point sensors and remote sensing pixels. This technique uses Distributed Temperature Sensing (DTS) over horizontal extents in the order of kilometers to infer soil moisture at this intermediate scale. Propagation of variations in atmospheric temperature and radiation with depth is a function of soil moisture. By using DTS temperature observations at three depths, it is possible to infer soil moisture assuming that heat conduction is the dominant heat transfer mechanism. The heat diffusion equation is inverted to obtain estimates of soil heat diffusivity and soil moisture. Since this technique relies on observations of the passive thermal response of the soil to atmospheric temperature and radiation variations rather than on sending active heat pulses, this technique is called passive SoilDTS in contrast to active soil DTS. The objective of Chapter 3 is to assess the feasibility of passive SoilDTS for soil moisture estimation. A field experiment was conducted in Monster, The Netherlands for this purpose.

The analysis of the experimental results presented in Chapter 3 points out a number of related technical and modeling issues that need to be investigated further in order for passive SoilDTS to be used for soil moisture estimation and scaling. Some soil moisture estimates were not reasonable due to uncertainties in cable depths and heat transfer mechanisms. To separate the technical issue of cable depth from the modeling challenges, the same methodology used to infer soil moisture from passive SoilDTS was applied to the profile data of temperature and soil moisture obtained with point sensors during the MicroWex field experiments in Florida in 2004.

(23)

depth. The objective of Chapter 4 is to obtain more insight into heat transfer mechanisms in the topsoil and the influence of soil moisture on these mechanisms.

In Chapter 4, the common assumption that conduction is the dominant heat transfer

mechanism in soil is clearly challenged. Differences between observations and predictions from a heat diffusion model are quantified as sink terms. The magnitude of the sinks indicates that evaporation plays a dominant role in heat transfer on dry days. Phase change has large impact on the heat budget of the soil. Compared to evaporation, transport of heat through liquid flow (percolation or capillary rise) is negligible and cannot explain the magnitude of the sinks. Yet evaporation rates are higher than would be expected if mass diffusion is the dominant transfer mechanism of water vapor. Vapor diffusion appears to be enhanced.

Chapter 5 focuses on understanding this enhanced vapor diffusion. In this Chapter, mechanisms that can explain enhanced vapor diffusion are reviewed and assessed with respect to the extent in which they can explain enhanced vapor diffusion in the topsoil. Order of magnitude of the mechanisms and their physical plausibility is investigated. From this analysis is concluded that enhanced evaporation is most likely caused firstly by lift of liquid water to the surface by small soil capillaries and secondary by enhanced mixing due to wind pumping. However, only wind pumping can also explain steep temperature profiles in the topsoil.

The hypothesis that wind pumping can cause enhanced vapor diffusion is investigated further in Chapter 6. Chapters 4, 5 and 6 deal with the third challenge addressed in this thesis, the modeling of the topsoil at a very fine depth resolution.

Finally, Chapter 7 presents a general discussion on the implications of this research for soil moisture retrieval and the assimilation of these retrievals in hydrological and meteorological models.

(24)

(25)

2

Comparing remotely sensed soil

moisture with precipitation: a case

study of the Iberian Peninsula

The research presented in this chapter was performed in the framework of the project “Needs Assessment of Future ESA Satellites for Water Management in Southern Europe” for the Netherlands Space Office. The results were presented at the European Geosciences Union Meeting in Vienna in 2009.

2.1. Introduction

Over the past decades, a range of remote sensing techniques have been developed to retrieve surface soil moisture from space-borne sensors. For water resources applications, both

operational and decision support, satellite remote sensing of soil moisture has advantages over ground-based measurements. The main advantages are spatial coverage and accessibility. Yet, soil moisture satellite remote sensing has limited applications in water resources management. This mainly stems from difficulties using the remotely sensed soil moisture in hydrological or water resources models. Modelers not only have difficulty assimilating the observations into a comparable model quantity, but also disaggregating the coarse scale observations to finer scales, as well as estimating the uncertainty of the observations.

Microwave sensors, both active scatterometers and passive microwave radiometers, are commonly used to estimate topsoil moisture from space. Active instruments send out short pulses towards the earth’s surface and measure the reflected pulse energy. The reflectivity of the soil increases with an increasing soil moisture content. Passive measurements are sensitive to soil moisture, because soil moisture influences the emissivity of the soil. For a more detailed description of soil moisture remote sensing, we refer to de Jeu et al. (2008).

Many efforts have been undertaken to validate remotely sensed soil moisture products.

Products have been compared to in situ measurements (e.g. Jackson et al., 2008), models (e.g. Scipal et al., 2008), other remotely sensed soil moisture products (e.g. Crow and Zhan, 2007), and data on related phenomena such as precipitation (e.g. Wagner et al., 2003; McCabe et al., 2005). Validation of soil moisture products is a difficult task, because of scale disparities and different possibilities in physical interpretation of the soil moisture estimates. Soil moisture is a state variable and therefore is the resultant of several interacting hydrological processes such as precipitation, evapotranspiration, runoff, drainage and groundwater processes. Near surface soil moisture retrievals from remote sensing are not the same as point measurements or depth averaged model estimates. It is critical to develop a variety of approaches to evaluate satellite retrieved soil moisture. One approach is through the identification of expected responses in related hydrological processes or proxy data streams such as precipitation (McCabe et al., 2005; Wagner et al., 2003).

(26)

In this chapter, we build on these approaches and use spatiotemporal precipitation patterns to validate and compare three different remotely sensed soil moisture products. Two hypotheses on the relationship between precipitation and soil moisture are tested. First, we will examine whether a certain temporal aggregation window of soil moisture reflects precipitation in space and second whether there is a direct response of soil moisture to precipitation.

2.2. Material and methods

2.2.1 Case study

The case study area is the Iberian Peninsula, which is a particularly interesting region due to its variable climate. The Iberian Peninsula, is located in the extreme southwest of Europe, and includes Spain, Portugal, Andorra and Gibraltar and a small part of France. It is bordered by the Mediterranean Sea and the Atlantic Ocean. The Pyrenees form the northeast edge of the Peninsula, separating it from the rest of Europe. The Iberian Peninsula has a surface area of 582,860 km², with elevations ranging from 0-3479 meters above sea level. It has a temperate climate with hot and dry summers in the interior and wetter and cooler summers along the coastlines, especially along the Atlantic coast. Large parts of the Iberian Peninsula are semi-arid. Annual precipitation ranges from up to 1000 mm over the Pyrenees and some areas in the northwest of the Peninsula to less than 300 mm in the semi-arid areas in the center and the southeast (see Figure 2.1).

Figure 2.1: Mean annual precipitation according to TRMM over the period from September 2003 to August 2006.

(27)

2.2.2 Data

In this study, three soil moisture products were compared. The selection of these three

products was based on their frequent use in both meteorological and hydrological applications. ERS Scatterometer

The active scatterometer is an instrument on board the ERS (1991 to present) and Metop satellites (>2006). It is operated in C-band (5.3 GHz) and measures the vertically polarized backscatter with three antennas at different azimuth and incidence angles. Measurements come at a spatial resolution 25-50 km with a temporal resolution of 3-35 days, depending on

competing data requests.

In order to retrieve soil moisture from backscatter observations, a change detection technique described in Wagner et al. (1999) is applied. Over land surfaces, backscatter measurements are affected by soil moisture, surface roughness and vegetation. The change detection technique exploits the temporal stability of surface roughness and vegetation compared to soil moisture in order to determine the soil moisture. Effects of vegetation phenology are further filtered out using the property that biomass affects backscatter differently at different incidence angles. The remaining signal is scaled to a degree of saturation, the surface soil water content (SWC), by relating the instantaneous backscatter measurement to a dry and wet backscatter reference. The reference values are determined by selecting the lowest and highest backscatter

measurements from a vegetation-corrected backscatter time series spanning a 10-year period. Results of this study are based on ERS2 scatterometer measurements obtained between September 1, 2003 and August 30, 2006. ERS data is available at a 30 km resolution and was interpolated to a 0.25° by 0.25° longitudinal latitudinal grid for this analysis.

AMSRENASA and AMSRENSIDC

The Advanced Microwave Scanning Radiometer - EOS (AMSR-E) is one of the six sensors aboard Aqua, which was launched in May 2002. AMSR-E is a passive microwave radiometer with the lowest frequency of 6.9 GHZ (C-band). The viewing angle of AMSR-E is a constant 55°. The spatial resolution of the measurements is approximately 50 km. We use two soil moisture products that use AMSR-E observations. The first product is a global soil moisture product developed by the Free University of Amsterdam and NASA (AMSRENASA; Owe et al., 2008). The

second AMSR-E product, is the official AMSR-E product developed at the United States National Snow and Ice Data Center (AMSRENSIDC;; Njoku et al., 2003).

The main difference between the two products can primarily be found in the use of different sensor frequencies, the method for modeling vegetation optical depth, and the estimation of soil/canopy temperature. The retrieval approach of the AMSRENASA product is based on the

microwave polarization difference index (Owe et al., 2001; de Jeu and Owe, 2003; Meesters et al., 2005). Dual polarized (H and V) observations are used to solve a radiative transfer equation for both soil moisture and vegetation optical depth. The VU-NASA algorithm requires no

calibration or auxiliary biophysical data and can be used with a range of microwave frequencies. The AMSRE C-band would probably give the best results, because sensitivity to soil moisture increases as frequency decreases. However, due to radio interference on the C-band, a slightly higher X-band (10.7 GHz) is often used for soil moisture retrieval over the study region. While the AMSRENSIDC (Njoku et al., 2003) uses the various responses of different channels to soil

moisture for retrieval.

As a consequence of the nature of the retrieval algorithms, soil moisture estimates have different units. Both AMSRENASA and AMSRENSIDC provide estimates for the volumetric water

content (VWC) of the topsoil. ERS provides estimates for the relative degree of saturation, denoted as soil water content (SWC). Information on soil characteristics, such as porosity and field capacity, is required in order to estimate the VWC from the SWC. This analysis focuses on relative changes rather than absolute values, so the difference between VWC and SWC is not relevant for this study. Hence, we will use the abbreviation SM for soil moisture.

(28)

In addition to the three soil moisture products, remote sensing data on precipitation was used in the analysis. Vegetation greenness and land cover maps were used to support our discussion of the results.

TRMM

The Tropical Rainfall Monitoring Mission (TRMM) is a joint project between NASA and the Japanese space agency, JAXA. TRMM was launched on November 27, 1997. The primary TRMM instruments are the Precipitation Radar (PR), the TRMM Microwave Imager (TMI) and the Visible Infrared Scanner (VIRS). TRMM processing algorithms combine information from these instruments and provides precipitation estimate at a scale of 0.25º x 0.25 º (Huffman et al., 2007). A range of orbital and gridded TRMM products are available. The 3 hourly- 3B42 is used for this study, aggregated to daily values.

NDVI and Land Cover Data

The VEGETATION instrument (VGT), on board the SPOT 4 and SPOT 5 satellites has four spectral bands: blue (0.43–0.47 μm), red (0.61–0.68 μm), Near InfraRed (NIR, 0.78–0.89 μm) and Short Wave InfraRed (SWIR, 1.58–1.74 μm). The red and NIR bands are used to calculate the Normalized Difference Vegetation Index according to the equation: NDVI=([NIR- RED]/[NIR + RED]). The spatial resolution of the imagery is 1 km. The NDVI data was scaled up to the 0.25° resolution of the other data sources using grid cell averaging. In this study, 10-day composites from the VGT2 sensor aboard the SPOT-5 satellite were used.

Land cover maps were taken from the CORINE (COoRdination of INformation on the

Environment) program at http://www.eea.europa.eu (Copyright EEA, Copenhagen, 2007). The original data was scaled up to 0.25° resolution by assigning each grid cell its dominant land use type.

2.2.3 Methods

The following two postulates are posited concerning the relation between topsoil moisture and precipitation:

1. For a certain temporal aggregation window, topsoil moisture resembles precipitation. Wagner et al. (2003) found that the correlation between ERS soil moisture and

precipitation was significant for large parts of the globe including the Iberian Peninsula. 2. A precipitation event above a certain threshold can lead to an increase in soil moisture.

McCabe et al. (2003) evaluated the concurrence of soil moisture increases and a few precipitation events over Iowa and found generally a good correspondence.

The first postulate is tested by comparing the three soil moisture products with TRMM using a correlation analysis. The correlation analysis was performed for aggregation windows of 1, 3, 5 and 10 days.

The ability of the soil moisture sets to capture precipitation events was tested by comparing the time series of precipitation occurrence with the first derivative of the soil moisture series. A precipitation occurrence was identified as exceeding a predefined threshold. For this analysis, we chose thresholds of 10, 20, 30 and 40 mm day-1_{. In order to avoid issues with overpass}

times, we defined an increase in soil moisture as follows: SM has increased the day after the precipitation event compared to the day before the precipitation event (ΔSM=SMt+1-SMt-1>0).

We used linearly interpolated soil moisture data in this analysis so that ΔSM could be calculated for all events.

The analysis is sensitive to overpass times, antecedent conditions and the fast drying of the shallow subsurface. Despite these factors, some concurrence between increases in soil moisture

(29)

Carlo experiments were performed with random signals. The generation of random precipitation signals was performed following Richardson (1980). The occurrence of precipitation was

modeled with a first order Markov chain. The magnitude of the events was taken from an exponential distribution. The Markov chain transition probabilities and the exponential

distribution parameter are seasonal for most of our study region. Parameters were derived from the time series for each pixel. The seasonal nature of these parameters was described by a second order Fourier series. The experiment on capturing precipitation was repeated for each pixel with 1000 randomly generated precipitation signals.

When evaluating the results of significance testing on a grid, the global or field significance should also be tested. For example, for an infinitely large set of independent series, 5 percent of the results can be locally significant at a 5 percent level by random chance and thus globally insignificant. If series are dependent in space, which is likely for soil moisture, the probability of finding significant pixels by random chance increases. By breaking the analysis into pixel-by-pixel regressions, multiplicity is introduced. A bootstrapping procedure to circumvent this issue has been used. Random precipitation time series were generated and the regression analysis repeated, determining the average regression coefficient. This process was then repeated 10,000 times and the actual results were compared to the bootstrapped distributions from the 10,000 synthetic regressions. Significance was assessed based on where the findings fall on the bootstrapped distribution. This procedure will also account for spatial autocorrelation in the precipitation data. See also: Barlow et al. (2001) as well as Wilks (1995).

2.3. Results and discussion

Figure 2.2 shows the correlation maps of the three soil moisture products and TRMM areas, where we found the highest correlations. These correlations were very different among the three maps. Correlations between ERS and TRMM are highest over the wetter parts of the Iberian Peninsula towards the northwest and the northeast (see the map in Figure 2.1 of TRMM precipitation for a comparison). Correlations between AMSRENASA and TRMM are highest

towards the southern coast and correlations between AMSRENSIDC and TRMM are highest for

some inland areas. The mean correlation over the grid (Table 2.1) is highest between ERS and TRMM. The mean correlation between the soil moisture product and precipitation increases, as expected, for an increasing aggregation window for AMSRENASA and AMSRENSDIC. The mean

correlation between ERS and TRMM remains approximately equal for all aggregation windows. This might be caused by a clear dominance of the seasonal cycle in ERS soil moisture and the sensors observation frequency. The average temporal resolution of ERS varies from 5 days over most of Portugal to more than 15 days over the Pyrenees.

Figure 2.2: Correlations of soil moisture and TRMM. Averaging window is 10 days for soil moisture and TRMM. Correlations range from 1 dark blue to -1 dark red.

(30)

Table 2.1: Results correlation analysis TRMM and SM.

Product 1 ERS AMSRENASA AMSRENSDIC

Product 2 TRMM TRMM TRMM Window (days) m ea n r 1 0.34 0.11 0.07 3 0.33 0.18 0.15 5 0.33 0.21 0.19 10 0.33 0.25 0.24 15 0.35 0.28 0.27

The analysis of the second postulate, concerning the ability to capture precipitation events, was performed for daily rain sums over 10, 20, 30 and 40 mm. Daily sums over 10 mm are found over the entire peninsula (see Figure 2.1). The number of days in which such events occur exceeds 50 over large areas of the peninsula. Daily sums over 40 mm are rarely found in the eastern part of the peninsula. AMSRENSIDC clearly shows best results in terms of captured

precipitation events (see Figure 2.3 and Table 2.2). The worst performance for this product was found over the Pyrenees and the northern coastal areas. AMSRENASA shows in general less

captured events than AMSRENSIDC. The Pyrenees and the northwestern part of the peninsula are

the areas where the statistics for this product shows poorer results. Compared to the two AMSR-E products, ERS shows the poorest results in capturing precipitation events. The poor results of ERS may be due to the low sensor’s observation frequency. The comparable METOP scatterometer has a higher observation frequency. Once a METOP time series is available, it can be used to test whether the low frequency is indeed the main reason for the scatterometer performance.

(31)

Figure 2.3: Captured precipitation events. The upper row shows the number of precipitation events over the respective thresholds 10 mm and 30 mm for the analysis period from September 2003 to August 2006. The other three rows show the fraction of total events that are captured for ERS, AMSRENASA and

AMSRENSIDC, respectively. The white pixels indicate no precipitation events over that magnitude.

(32)

The mean of captured events over the grid generally increases as the size of the event increases (except for AMSRENASA and the threshold of 40 mm). This seems logical, since with

larger events the probability that the soil will dry out between two subsequent satellite observations is lower than for smaller events. Larger events also have a lower probability of having been preceded by an even larger event which would have had a dominant effect on soil moisture changes.

The significance testing (Table 2.2) confirms results described in the previous paragraph. Precipitation events are best captured by AMSRENSDIC, followed by AMSRENASA, while the ERS

gives the poorest results. As the precipitation threshold increases the significance of the results decreases, even though the mean of the captured events over the grid increases. This can be explained by a decreasing number of precipitation events for increasing thresholds.

Table 2.2: Summary of the results of the test of captured precipitation events. The third column shows the mean of the captured events divided by the total of events over the given threshold. The fourth and fifth columns show the number pixels in which the locally significant number of precipitation events captured (see Figure 2.1) is divided by the total number of pixels where the events occurred that were over the given threshold.

Product TRMM threshold (mm) Mean captured pixels p<0.1 pixels p<0.05 ERS 10 0.56 0.47 0.34 20 0.59 0.33 0.18 30 0.60 0.14 0.05 40 0.63 0.02 0.00 AMSRENASA 10 0.64 0.81 0.75 20 0.66 0.60 0.44 30 0.66 0.20 0.09 40 0.64 0.05 0.01 AMSRENSIDC 10 0.72 0.88 0.84 20 0.75 0.78 0.68 30 0.78 0.38 0.22 40 0.81 0.08 0.02

It is important when evaluating the results of significance testing on a grid, that the global or field significance should also be tested. For an infinitely large set of independent series, 5 percent of the results can be locally significant at a 5 percent level by random chance and therefore globally insignificant. If the series are dependent in space, which is likely the case for soil moisture, then the probability of finding significant pixels by random chance increases. By using a simple binominal test described in Wilks (1995), we found that for precipitation events over 30 and 40 mm, the results for the ERS were not significant at the global level. For

AMSRENASA and AMSRENSIDC, this is the case for precipitation events over 40 mm. The binominal

test assumes an independence between the analyzed series, which is not a valid assumption for soil moisture. More accurate random signals could have been obtained with Weibul distribution (Selker & Haith, 1990) instead of the first order Markov Chain used here. Yet due to large sample sizes (a minimum of 8000 pixels depending on the chosen precipitation threshold) and the fact that the results are far away from the critical limits, more sophisticated tests were not considered necessary.

We compared the results with vegetation and land use maps. Correlations between the ERS and TRMM were generally highest for those areas with high precipitation (Figure 2.2), high NDVI (Figure 2.4) and land use type scrubs, forest or heterogeneous agriculture (Figure 2.4). This is counterintuitive as we would expect microwave remote sensing to perform better over sparsely

(33)

ERS to vegetation water content as suggested by Friesen et al. (2012). Capturing of

precipitation events is generally better over sparsely vegetated areas. Notably, AMSRENASA but

also AMSRENSIDC capture precipitation events (Figure 2.2) better over areas with low NDVI

(Figure 2.4). The differences between the ERS and both AMSRE products in capturing precipitation events was the largest over arable land (Figure 2.4).

Figure 2.4: Mean NDVI and land use according to the second level of the Corine database.

This research confirms the importance of different evaluation strategies of soil moisture as argued by McCabe et al. (2004). A correlation analysis between the three soil moisture products (not shown) yields the best results for comparing the ERS with the AMSRENASA. For the western

part of the Iberian Peninsula correlations (r) between ERS and AMSRENASA are over 0.5, whereas

the correlations between AMSRENSDIC and the other two products hardly reach this value. Yet in

capturing precipitation events, AMSRENSDIC clearly outperforms the other two. The three

evaluated soil moisture datasets show very different response both in time and space that has direct implications for using the data for water resources applications. Due to the large difference in their results, the three moisture products should not simply and interchangeably be assimilated as topsoil moisture in a model.

2.4. Conclusion

Over the Iberian Peninsula, the performance of three remotely sensed soil moisture products weres compared by evaluating two postulates on the relationship between precipitation and soil moisture. The postulates were: (1) that for a certain temporal aggregation window, soil

moisture reflects precipitation; and (2) that there is concurrence between the increase in soil moisture and precipitation events.

Results for the three soil moisture datasets, ERS, AMSRENASA and AMSRENSIDC show very

different response in time and in space. While the time series correlation for precipitation was on average highest for the ERS, AMSRENSIDC performed best in capturing precipitation events.

The results stress the importance of using different methods to comparing soil moisture products. Due to the large temporal and spatial variability between soil moisture products, these products should not be used interchangeable in modeling studies and data assimilation. More research into the nature of these differences is still required.

(34)

(35)

3

Feasibility of soil moisture

estimation using passive

distributed

temperature sensing

This chapter is based on Steele-Dunne et al. (2010).

3.1. Introduction

There are large differences between soil moisture products as demonstrated in the previous chapter. Retrieval algorithms are far from converged and need to be improved. In order to improve the calibration and validation of remote sensing soil moisture products, a global network of independent in situ soil moisture measurements over varying soil and land cover types is essential. Point observations of soil moisture are relatively straightforward to make use of established methods such as a time domain reflectometry (TDR) and gravimetric sampling. TDR is a useful method for making continuous, long‐term measurements since it is

nondestructive and sensors can be installed at the site of interest. However, monitoring large‐ scale variability with TDR would involve installing a vast and costly network of sensors. Here we propose using distributed temperature sensing (DTS) to obtain simultaneous measurements of soil moisture over large areas. By providing continuous, high-resolution observations over a large area, soil DTS could play an important role in supporting a modest network of traditional sensors. Soil moisture fields can maintain spatial patterns in time because of covariances between soil moisture and factors such as topography, soil texture, and

vegetation (Vachaud et al., 1985; Mohanty and Skaggs, 2001; Jacobs et al., 2004; Cosh et al., 2004). Temporal and spatial stability concepts can be used to identify a single or a few representative sensor locations, observations which are similar to the field or pixel average. Alternatively, they can also be used to identify locations that are systematically biased with respect to the mean, thereby providing a measure of subfield or subpixel heterogeneity. Quantifying spatiotemporal stability requires an extensive initial network of sensors to measure soil moisture over a lengthy validation period. Soil DTS offers a relatively economical way to make continuous observations at thousands of locations within the watershed or footprint of interest. Furthermore, combining a few accurate, conventional sensors with the distributed observations from soil DTS to capture fine‐scale variability in soil moisture would enhance the usefulness of large‐scale remote sensing products.

The idea of observing temperature dynamics to infer soil moisture is far from new. Idso et al. (1975a, 1975b, 1976) explored the possibility of using the thermal inertia of the surface to infer

Moisture in the topsoil: From large-scale observations to small-scale process understanding