robotic ecological mapping: habitats and the search for life in the atacama desert

Robotic ecological mapping: Habitats and the search for life in the

Atacama Desert

K. Warren-Rhodes,1 S. Weinstein,2 J. L. Piatek,3 J. Dohm,4 A. Hock,5 E. Minkley,2

D. Pane,2 L. A. Ernst,2 G. Fisher,2 S. Emani,2 A. S. Waggoner,2 N. A. Cabrol,6

D. S. Wettergreen,7 E. Grin,6 P. Coppin,8 Chong Diaz,9 J. Moersch,3 G. G. Oril,7

T. Smith,7 K. Stubbs,7 G. Thomas,10 M. Wagner,7 M. Wyatt,4 and L. Ng Boyle10

Received 6 September 2006; revised 26 January 2007; accepted 17 May 2007; published 25 September 2007.

[1] As part of the three-year ‘Life in the Atacama’ (LITA) project, plant and microbialabundance were mapped within three sites in the Atacama Desert, Chile, using anautomated robotic rover. On-board fluorescence imaging of six biological signatures (e.g.,chlorophyll, DNA, proteins) was used to assess abundance, based on a percentpositive sample rating system and standardized robotic ecological transects. The percentpositive rating system scored each sample based on the measured signal strength (0 for nosignal to 2 for strong signal) for each biological signature relative to the total ratingpossible. The 2005 field experiment results show that percent positive ratings variedsignificantly across Site D (coastal site with fog), with patchy zones of high abundancecorrelated with orbital and microscale habitat types (heaved surface crust and gravel bars);alluvial fan habitats generally had lower abundance. Non-random multi-scalebiological patchiness also characterized interior desert Sites E and F, with relatively highabundance associated with (paleo)aqueous habitats such as playas. Localized variables,including topography, played an important, albeit complex, role in microbialspatial distribution. Site D biosignature trends correlated with culturable soil bacteria, withMPN ranging from 10-1000 CFU/g-soil, and chlorophyll ratings accurately mappedlichen/moss abundance (Site D) and higher plant (Site F) distributions. Climate alsoaffected biological patchiness, with significant correlation shown between abundance and(rover) air relative humidity, while lichen patterns were linked to the presence of fog.Rover biological mapping results across sites parallel longitudinal W-E wet/dry/wetAtacama climate trends. Overall, the study highlights the success of targeting of aqueous-associated habitats identifiable from orbital geology and mineralogy. The LITA experiencealso suggests the terrestrial study of life and its distribution, particularly the fields oflandscape ecology and ecohydrology, hold critical lessons for the search for life on otherplanets. Their applications to robotic sampling strategies on Mars should be furtherexploited.

Citation: Warren-Rhodes, K., et al. (2007), Robotic ecological mapping: Habitats and the search for life in the Atacama Desert,

J. Geophys. Res., 112, G04S06, doi:10.1029/2006JG000301.

1. Introduction

[2] Important strides in the robotic search for microbiallife in hyperarid deserts were made during the ‘Life in theAtacama’ (LITA) project’s remote science field experiment

in 2004 in the Atacama Desert, Chile [Cabrol et al., 2007b].The LITA study’s overall science and technical objectiveswere to locate microbial life and map habitats in severalclimatic regions (coastal, interior) of the Atacama Desert(Figure 1), and in 2004 two sites (B and C) were

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 112, G04S06, doi:10.1029/2006JG000301, 2007ClickHere

for

FullArticle

1NASA Ames Research Center, Moffett Field, California, USA.2Molecular Biosensor and Imaging Center, Carnegie Mellon University,

Pittsburgh, Pennsylvania, USA.3Department of Earth and Planetary Sciences, University of Tennessee,

Knoxville, Tennessee, USA.4Department of Hydrology and Water Resources, University of Arizona,

Tucson, Arizona, USA.5Department of Earth and Space Sciences, University of California, Los

Angeles, California, USA.

Copyright 2007 by the American Geophysical Union.0148-0227/07/2006JG000301$09.00

G04S06

6SETI Institute, Mountain View, California, USA.7Robotics Institute, Carnegie Mellon University, Pittsburgh, Pennsyl-

vania, USA.8Eventscope, Carnegie Mellon University, Pittsburgh, Pennsylvania,

USA.9Universidad Catolica del Norte, Antofagasta, Chile.10Department of Mechanical and Industrial Engineering, University of

Iowa, Iowa City, Iowa, USA.

1 of 16

http://dx.doi.org/10.1029/2006JG000301

investigated. An overview of the biological results for thisfirst field study is found in S. Weinstein et al. (Applica-tion of pulsed-excitation fluorescence imager for daylightdetection of sparse life in tests in the Atacama Desert,submitted to Journal of Geophysical Research, 2007,hereinafter referred to as Weinstein et al., submittedmanuscript, 2007) and Warren-Rhodes et al. [2007a].Based on the LITA 2004 results, a major objective ofthe second field campaign held in 2005 was to integratestandard terrestrial ecology methods with robotic sam-pling to facilitate in-situ hypothesis testing and post-traverse comparisons. In 2005, three sites in the Atacamawere investigated (Figure 1): (1) Site D, a ‘wet’ coastallocation (�10 km2 study area); (2) Site E, an extremely‘dry’ inland location (�40 km2) in the Atacama’s hyper-arid core; and (3) Site F, an area (�40 km2) in theextremely dry interior situated closer (relative to site E) tothe Andes.[3] During the LITA study, a team of scientists (biolo-

gists, ecologists, geologists and spectroscopists) workedremotely from Carnegie Mellon University as an automatedrover surveyed field sites in-situ in the Atacama Desert.Throughout the project, science team members had noknowledge of the whereabouts of any particular LITA fieldlocation [Cabrol et al., 2007b]. Each day (sol), data col-lected by the rover in the Atacama was downloaded,analyzed and used by the remote science team to assessthe environment surveyed, generate and test hypotheses and

build the subsequent day’s sampling plan. Individual sitestudies lasted �1 week [Cabrol et al., 2007b], and at theproject’s conclusion, a post-operations (post-ops) ground-truth survey (January 2006) was completed. This paperprovides a detailed review of the 2005 results, with aparticular focus on Site D which served as a test case forrobotic mapping with transects, and it highlights findingsfrom other locations in the 2005 field experiment toillustrate intra-site differences and overall conclusions.Additional information on the LITA project and scienceresults is provided elsewhere [Cabrol et al., 2007b], includ-ing the 2004 biological and ecological results [Warren-Rhodes et al., 2007a], rover instrumentation, microbiologyand fluorescence detection methods (Weinstein et al., sub-mitted manuscript, 2007), and mineralogy (J. Piatek et al.,Surface and subsurface composition of the Life in theAtacama field sites from rover data and orbital imageanalysis, submitted to Journal of Geophysical Research,2007, hereinafter referred to as Piatek et al., submittedmanuscript, 2007).

2. Sampling Strategy and Instrumentation

[4] A cornerstone of many ecological studies is anestimate of the abundance and diversity of populations[Krebs, 1998]. Decades of research have yielded standardtechniques geared towards ecological study design, repre-sentative sampling, and statistical analysis of plant andanimal populations [Begon et al., 1990; Krebs, 1998; Dale,1999; Turner et al., 2001]. In contrast, microbial communitysampling continues to vary widely in terms of techniques,application and results [e.g., Nunan et al., 2002, 2003;Grundmann, 2004; Green and Bohannan, 2006]. Giventhe lack of precedent coupled with robotic engineeringconstraints, the specific sampling approach chosen for the2005 LITA field study was necessarily experimental and itsexecution imperfect. Nonetheless, the results shed newinsights into the distribution of life in one of the world’sdriest deserts and provide a first step towards future stan-dardized rover biological studies.

2.1. Selection of Site D (Coastal) Regions of BiologicalInterest From Satellite Imagery

[5] Prior to the rover traverse, regions of biologicalinterest within the rover’s known ‘landing ellipse’(Figure 2) were selected from satellite imagery basedon several astrobiological criteria: (1) mineralogy (e.g.,presence of sulfates, carbonates, quartz); (2) past or presentaqueous activity (e.g., basins, fog-impacted areas)—i.e., a‘‘follow-the-water’’ strategy; (3) habitats (e.g., hypersaline)and (4) biosignatures (e.g., chlorophyll). (For a detaileddescription of this process to select regions of interest, seeWarren-Rhodes et al. [2007a].) Figure 2 shows the Site Dregion of interest results, with the western half of thelanding ellipse hypothesized by the science team (prior tothe traverse) to contain high priority regions for biology dueto ancient or recent water activity, including: (1) fog-proneareas near gaps in the Pacific coastal range (visiblefrom orbit); and (2) topographic lows (e.g., basins).The latter areas were also a priority based on increasedsulfate concentrations observed in satellite images, poten-tially associated with runoff. No chlorophyll signatures

Figure 1. LITA site locations (adapted from Riquelme etal. [2003]) within major longitudinal physio-geographicalunits (coastal to Andean) in the Atacama Desert, Chile.

G04S06 WARREN-RHODES ET AL.: ROBOTIC ECOLOGICAL MAPPING IN ATACAMA

2 of 16

G04S06

were detected in satellite spectra (Piatek et al., submittedmanuscript, 2007).

2.2. Robotic Ecological Transects: The StandardPeriodic Sampling Unit

[6] The rover-mounted microscopic and fluorescenceimager (FI) detects microbial (and plant) communities viatargeted fluorescent probes for DNA, protein, carbohydratesand lipids—the biomolecules contained within organisms’cells and external structures (sheaths, biofilms, cell walls)—and via chlorophyll directly, which is naturally possessed byplants, lichens and certain photosynthetic microbes (Wein-stein et al., submitted manuscript, 2007). Each full FIsample—defined as the completion of all biosignature tests,i.e., visible, chlorophyll, DNA, and protein (in 2004) pluscarbohydrates and lipids (in 2005)—required 35 minutes(Weinstein et al., submitted manuscript, 2007).[7] Tradeoffs existed between biology and rover mobility

(distance for reconnaissance) that limited the total number

of daily FI samples [Cabrol et al., 2007b]. This tradeoffprovided impetus to maximize datapoints and devise sam-pling strategies that would facilitate hypothesis testingwithin and across sites. Ideally, to estimate abundance andspatial distribution, a standard length sampling unit wouldcollect comparable data across multiple scales—e.g., a 1 m-by�100 m transect within which 5 full samples are acquired(i.e., every 20 m), with n > 1 replicate transects per area. Inpractice, engineering specifications dictated a 10-cm sam-pling unit width (FI field of view), a 30 m distance betweensamples (resolution of a DEM pixel) and an �180 m totaltransect length (based on available rover operation time andbandwidth constraints). Autonomous rover navigation alsoprecluded strict linear transects, and time constraints per-mitted a full sample to be taken only at the beginning andend of a transect. In contrast, samples between these firstand last points were defined as ‘‘periodic’’ and exploited therover’s ‘‘Science on the Fly’’ (SOTF) capabilities [seeCabrol et al., 2007b]. During SOTF mode, a positive signal

Figure 2. ASTERVNIR single band image (15 m/pixel) of Site D Regions of Interest Map. Geologicalunits are plains-forming units (p0, p1 and p3), e2 (etched unit) and k (knobby unit). Light blue dotsindicate regions likely to be influenced by atmospheric water vapor, and blue arrows suggest thewestward flow of marine fog through gaps in the Pacific coastal range. Dark blue dots indicate regionsshaped by past or present aqueous activity, including topographic depressions. Green dots representregions chosen from satellite spectra for high potential for life based on relatively high concentrations ofsulfate minerals. Light albedo regions are also suggestive of sulfate concentrations, but are not definitiveindicators of evaporite mineralogy, and were not specifically targeted.


3 of 16

G04S06

obtained for an indicator channel—chosen as the chloro-phyll channel to save time (�10 minutes per sample)—triggers a rover stop to collect a full sample, while anegative signal instructs the rover to continue the transect.By definition, any location where one or more samplinginstruments was deployed along the rover traverse wasdesignated a ‘locale.’[8] The LITA ‘‘standard periodic sampling unit’’ (SPSU)

was thus designed as an 180 m transect, with 2 full samples(one at each end of the transect) and 5 periodic samples(every �30 m, visible and chlorophyll data acquired only,unless a ‘stop’ was triggered by a positive chlorophyllsignal). In reality, the SPSU typically consisted of an�180–210 m (n = 6–8 periodic samples) transect, withoutliers of �150–270 m. Table 1 presents details for Site Dsamples and SPSU.[9] Along with satellite imagery and ground-based bio-

logical data, a wealth of additional data (Table 1) wascollected by the rover’s instruments, including a visible/near-infrared (VNIR) and a thermal infrared (TIR) spec-trometer (human-operated) that collected data on mineralogy(see Piatek et al. (submitted manuscript, 2007) and Warren-Rhodes et al. [2007a] for ground-based VNIR and TIRresults); workspace and stereo panoramic (SPI) cameras formacroscale and microscale imagery that furnished contex-tual ecology and geology (km to cm); and on-boardenvironmental sensors, including an Onset Hobo Pro

1

datalogger for air temperature and relative humidity(RHA) and a Campbell leaf wetness sensor to detectcondensed liquid water from fog or dew. Additionally,meteorological data, including solar insolation (pyronome-ter), wind speed and direction, and air temperature andrelative humidity, were also collected by a Campbell

1

weather station installed for the duration of the fieldexperiment at the site’s first sampling location. Lastly,subsurface soil habitats were accessed by a plow mountedbelow the rover that removed �1 cm of surface soils. Fordetailed rover payload descriptions, see Cabrol et al.[2007b].

3. Results

3.1. Within Site Analyses, Test Case of Site D

3.1.1. Overall Site D Results[10] During the 14 km Site D rover science traverse,

26 distinct locales, 9 SPSU and 92 individual FI sampleswere completed. Of these samples (Table 2), 30 were fullsamples (11 triggered during SOTF periodic sampling) and62 were visible/chlorophyll samples only (hereafter‘‘CHL,’’ with most being periodic samples). Ninety percentof full samples (hereafter FS) were ‘‘positive’’ (defined asany positive signal in �1 biosignature channel) versus 30%for CHL, indicating a significantly higher detection successwith FS (all biosignatures) versus CHL only.[11] In addition to this simple definition and rating scheme

for a positive sample, a more specific and conservativepercent positive biosignature rating system (a proxy forrelative microbial abundance) was also devised. Each ofthe six biosignature channels [visible, chlorophyll, DNA,protein, carbohydrates and lipids] was rated from 0 (nopositive signal) to 2 (strong positive signal), for a maximum12 points total. If all 6 channels exhibited strong positive

signals, the sample’s total percent positive biosignaturerating was 100% (Table 1). [Note that for 2004, only thefirst four biosignature channels were used.] In addition,results could also be examined based on CHL (indicatingphotosynthetic organisms) and/or the DNA-Protein-Lipid-Carbohydrate channels only (DPLC). Mean percent positiveratings for individual sites (mean of the percent positiveratings for all samples within the site) or SPSU (mean ofpercent positive ratings for all samples comprising theSPSU) were also calculated. Hereafter, all reference tomicrobial abundance, unless otherwise noted, refers to eitherthe simple percent scheme or a total percent positivebiosignature rating (of an individual sample, meanSPSU±standard error (SE), and/or the meansite ±SE).3.1.2. Site D (Coastal Location) Traverse HabitatOverview[12] Despite a homogeneous mineralogical environment

at Site D (Piatek et al., submitted manuscript, 2007),significant diversity at the orbital, ground-based and micro-scopic scale existed in geology and ecology (Table 1),producing concomitant variations in habitats and microbialabundance, spatial distribution and diversity (Figure 3). (Fora definition and explanation of habitat types, see Warren-Rhodes et al. [2007a].) Multi-scale habitat analysis (satel-lite, ground-based SPI, microscopic FI) revealed three mainhabitat types (Figure 3): (1) desert pavement, i.e., soilsmantled by gravels; (2) heaved/raised surface crust, oftenappearing porous and in certain locales associated withsulfates, which have a high capacity for water absorptionand thus microbial colonization—hereafter ‘‘surface crusthabitats’’; and (3) gravel bars. Table 3 contains a shorttraverse overview, including the locales surveyed, habitattypes, transitions and main observations.3.1.3. Site D Traverse Biological Results[13] During its traverse and mapping of Site D, the rover

measured significant variations in microbial abundance anddiversity. Of 92 FI samples (Table 1), half were positive(positive signal in any channel), with the meanall samples =30 ± 4% (Table 2). CHL results were positive for 29% ofsamples, and these signals were visibly correlated withlichens and/or moss (Figure 3) (see also fluorescenceimages in Weinstein et al. (submitted manuscript, 2007)).A diversity of lichen types [assumed from morphology andcolor (grey, black, white, yellow, orange)] was observedwithin a single FI sample/locale, and variation in diversitybetween FI samples/locales also existed. Moss distributionwas more restricted than lichens. In contrast to CHL results,positive DPLC ratings (i.e., excluding chlorophyll-basedpositive signals) were obtained in 52% of the full samplesacquired at Site D (detailed data not shown in Table 1).[14] Most importantly, from FI results, the rover mapped

4 macroscale ‘‘zones of high abundance’’ (Figure 3). SPSUwithin these ‘‘biological hotspots’’ had significantly higherbiosignature ratings (Figure 4) than other SPSU [F-test,df = 8, 61; F = 7.83, P < 0.0001]. Zones of markedly lowrelative abundance were also mapped (Figures 3 and 4).Differences in least square means indicate that SPSU 2 (meanpercent positive ratingSPSU2 = 5.25 ± 10%) was significantlydifferent from SPSU 1, 3, 6 and 7 (with 7 having the largestmean percent positive at 86 ± 10%) (Figure 4). Other differ-ences in SPSU are presented in Table 4. This extremeheterogeneity, i.e., patchiness, in microbial abundance and


4 of 16

G04S06

Table 1. General Features of Rover-Acquired Samples, Site D

SPSUa Sol FI Id Tagb

% PositiveBio-

Signaturesc %RHdDistance

to Gap (m)eHabitat(Orbital)f

Habitat(Panorama)g

Habitat(Micro)h

2 locale 10 034 30 95 5300 M, AF DP/wash d22 locale 10 42 (plow) 0 95 5300 M, AF DP/wash d22 locale 10 43 (plow) 0 95 5300 M, AF DP/wash d22 locale 10 49 (plow) 0 95 5300 M, AF DP/wash d22 locale 10 50 (plow) 0 95 5300 M, AF DP/wash d22 locale 10 65 (plow) 0 95 5300 M, AF DP/wash d22 locale 10 66 (plow) 0 95 5300 M, AF DP/wash d22 locale 10 68 (plow) 0 95 5300 M, AF DP/wash d22 locale 10 69 (plow) 0 95 5300 M, AF DP/wash d22 locale 20 071 TRN wypt 30 0 95 5300 M, AF DP/wash d32 locale 30 074 TRN wypt 075 0 95 5300 M, AF DP/wash d22 locale 40 077 TRN wypt 130 100 95 5300 M hillslope Heaves2 locale 50 080 TRN wypt 262 BD 100 5300 M hillslope Heaves2 locale 50 081 20 100 5300 M hillslope Heaves

1 3 locale 50 127 42 100 5300 M hillslope Heaves1 3 locale 60 116 TRN wypt 358 100 100 5300 M hillslope Heaves1 3 locale 60 116 TRN wypt 384 25 100 5300 M hillslope Heaves1 3 locale 60 116 TRN wypt 409 0 100 5300 M hillslope Heaves1 3 locale 60 116 TRN wypt 434 0 100 5300 M hillslope Heaves1 3 locale 60 116 TRN wypt 460 0 100 5300 M hillslope Heaves1 3 locale 60 116 TRN wypt 485 25 100 5300 M hillslope Heaves1 3 locale 60 141 83 100 5300 M hillslope Heaves

3 locale 90 002 runout FI BD 80 2296 AF DP d32 4 locale 90 127 BD 80 2296 AF DP d32 4 locale 100 206 4370 0 na 2296 AF DP d32 4 locale 100 206 4395 0 na 2296 AF DP d32 4 locale 100 206 4420 0 na 2296 AF DP d22 4 locale 100 206 4445 0 na 2296 AF DP d22 4 locale 100 206 4470 0 na 2296 AF DP d22 4 locale 100 206 4495 25 na 2296 AF DP d22 4 locale 100 206 4520 0 na 2296 AF DP d22 4 locale 100 223 17 na 2296 AF DP d3

4 locale 110 217 (plow) 25 na 1306 AF/Confluence DP/gullies d34 locale 110 218 (plow) 100 na 1306 AF/Confluence DP/gullies d3

3 4 locale 110 212 83 na 1306 AF/Confluence DP/gullies d33 4 locale 120 208 TRN 5531 25 97 1306 AF/Confluence DP/gullies d33 4 locale 120 208 TRN 5563 100 97 1306 AF/Confluence DP/gullies d33 4 locale 120 208 TRN 5588 100 97 1306 AF/Confluence DP/gullies d33 4 locale 120 208 TRN 5613 0 97 1306 AF/Confluence DP/gullies d33 4 locale 120 208 TRN 5638 50 97 1306 AF/Confluence DP/gullies d33 4 locale 120 208 TRN 5663 0 97 1306 AF/Confluence DP/gullies d23 4 locale 120 222 33 97 1306 AF/Confluence DP/gullies d2

4 locale 121 002 0 97 1659 AF-middle DP/gullies d25 locale 121 267 0 97 1659 AF-middle DP/gullies d25 locale 130 270 TRN 6015 BD na 1659 AF-middle DP/gullies d35 locale 130 270 TRN 6040 BD na 1659 AF-middle DP/gullies d35 locale 130 270 TRN 6066 BD na 1659 AF-middle DP/gullies d35 locale 130 270 TRN 6091 BD na 1659 AF-middle DP/gullies d35 locale 130 270 TRN 6116 BD na 1659 AF-middle DP/gullies d35 locale 130 270 TRN 6141 BD na 1659 AF-middle DP/gullies d35 locale 130 270 TRN 6166 BD na 1659 AF-middle DP/gullies d3

4 5 locale 130 271 0 na 1659 AF-middle DP/gullies d34 5 locale 140 275 TRN 6202 25 na 1574 AF-upper DP/gullies d34 5 locale 140 275 TRN 6227 0 na 1574 AF-upper DP/gullies d34 5 locale 140 275 TRN 6252 0 na 1574 AF-upper DP/gullies d34 5 locale 140 275 TRN 6277 0 na 1574 AF-upper DP/gullies d34 5 locale 140 275 TRN 6302 0 na 1574 AF-upper DP/gullies d34 5 locale 140 275 TRN 6327 0 na 1574 AF-upper DP/gullies d34 5 locale 140 275 TRN 6352 0 na 1574 AF-upper DP/gullies d34 5 locale 140 275 TRN 6377 0 na 1574 AF-upper DP/gullies d14 5 locale 140 276 33 na 1574 AF-upper DP/gullies d15 5 locale 190 296 33 na 2803 sulfate hillslope hillslope Heaves5 5 locale 200 TRN 9354 0 na 2803 sulfate hillslope hillslope Heaves5 5 locale 200 TRN 9379 0 na 2803 sulfate hillslope hillslope Heaves5 5 locale 200 TRN 9404 0 na 2803 sulfate hillslope hillslope Heaves5 5 locale 200 TRN 9430 0 na 2803 sulfate hillslope hillslope Heaves5 5 locale 200 002 BD 80 2803 sulfate hillslope hillslope Heaves6 6 locale 200 365 67 80 2803 sulfate hillslope hillslope Heaves6 6 locale 210 328 TRN 9618 100 80 2803 sulfate hillslope hillslope Heaves


5 of 16

G04S06

diversity (reflected by distinct hotspots) was observed atboth the macroscale (km), and also across and within SPSU(m to cm scales). Within SPSU 1, for example, total percentpositive ratings for individual samples ranged from 0% to100% (Table 1), with the first three samples in the SPSUhaving relatively high ratings, while the next three sampleswere negative (0%). Concomitant with biological dataanalysis, the science team performed detailed habitat map-ping (Tables 1 and 2 and Figure 3) during the fieldexperiment using multi-scalar rover-acquired (FI/SPI/VNIR) and satellite datasets. Simultaneous habitat mappingduring the rover traverse (1) facilitated the generation ofhypotheses to explain the observed biological patchiness,and (2) assisted the science team in planning for andcollecting necessary data to test hypotheses and to identifyspecific factors [section 3.2] that controlled this biologicalheterogeneity.

3.1.4. Site D Field Ground-Truth and LaboratoryResults[15] Overall, post-ops ground-truth supported both the

microscale and macroscale conclusions (e.g., microbialpatchiness) derived by the science team from satellite androver ground-based data. However, post-ops ground truthenabled much more definitive insights to be gleaned as tothe explanations underlying certain phenomena mapped bythe rover and/or the linkages between certain factors (e.g.,topography) and relative microbial abundance and spatialdistribution.[16] Specifically, in terms of the Site D traverse, ground-

truth observations confirmed the first habitat transitionmapped by the rover (Figure 3 and Table 3) at the geolog-ical contact (�locale 40) separating pavement/wash habitat(locales 10–30) with low but patchy microbial abundancefrom heaved surface crust with high abundance (locales

SPSUa Sol FI Id Tagb

% PositiveBio-

Signaturesc %RHdDistance

to Gap (m)eHabitat(Orbital)f

Habitat(Panorama)g

Habitat(Micro)h

6 6 locale 210 328 TRN 9644 100 80 2803 sulfate hillslope hillslope Heaves6 6 locale 210 328 TRN 9670 58 80 2803 sulfate hillslope hillslope Heaves6 6 locale 210 328 TRN 9696 0 80 2803 sulfate hillslope hillslope Heaves6 6 locale 210 329 40 80 2803 sulfate hillslope hillslope Heaves

6 locale 240 392 100 na 3306 valley floor DP Heaves7 6 locale 250 378 67 na 3306 valley floor DP Heaves7 6 locale 260 383 TRN 10891 83 na 3306 valley floor DP Heaves7 6 locale 260 383 TRN 10917 83 na 3306 valley floor DP Heaves7 6 locale 260 383 TRN 10943 75 na 3306 valley floor DP Heaves7 6 locale 260 383 TRN 10969 100 na 3306 valley floor DP Heaves7 6 locale 260 383 TRN 10995 100 na 3306 valley floor DP Heaves7 6 locale 260 383 TRN 11020 100 na 3306 valley floor DP Heaves7 6 locale 260 384 80 100 3306 valley floor DP Heaves8 7 locale 260 422 17 100 3948 AF exposed crust d28 7 locale 270 430 TRN 11612 0 100 3948 AF exposed crust d38 7 locale 270 430 TRN 11637 0 100 3948 AF exposed crust d38 7 locale 270 430 TRN 11662 0 100 3948 AF exposed crust d38 7 locale 270 430 TRN 11687 83 100 3948 AF exposed crust d18 7 locale 270 430 TRN 11714 40 100 3948 AF exposed crust d38 7 locale 270 431 100 100 3948 AF exposed crust d39 7 locale 290 445 50 100 1958 AF DP d19 7 locale 300 449 TRN 14112 0 100 1958 AF DP d19 7 locale 300 449 TRN 14138 58 100 1958 AF DP d29 7 locale 300 449 TRN 14164 0 100 1958 AF DP d29 7 locale 300 449 TRN 14189 0 100 1958 AF DP d29 7 locale 300 449 TRN 14214 0 100 1958 AF DP d29 7 locale 300 449 TRN 14240 0 100 1958 AF DP d29 7 locale 300 449 TRN 14265 0 100 1958 AF DP d29 7 locale 300 449 TRN 14289 0 100 1958 AF DP d29 7 locale 300 450 0 100 1958 AF DP d1aSPSU = standard periodic sampling unit, which was an �180 m transect with a full FI sample (visible, chlorophyll, DNA, protein, carbohydrate, and

lipids) at the first (0 m) and last locale (�180 m) and periodic waypoint FI samples (visible and chlorophyll) spaced every �30 m.bSample Identification Tag: Name of sample acquired by fluorescence imager; bold sample id tags indicate a habitat transition was identified during the

science traverse, based on SPI and FI data. Plow = rover plow was deployed to remove �top 1 cm of soil surface.cPercent Positive Biosignatures: Sample abundance was measured as = sample positive rating/total possible rating X 100. For a full FI sample, the total

maximum rating was 12 [2 points possible for each biosignature category (visible, chlorophyll, DNA, protein, lipids, and carbohydrates) and for a periodicwaypoint, 4 points. BD = bad data (e.g., blurry), which was subtracted from total points possible]; bold here indicates a full sample, with all others periodic.

dPercent air relative humidity on-board the rover at a locale, with �95% suggestive of the presence of fog (but confirmed by other data, such as SPIimagery, leaf wetness and solar insolation data).

eDistance from coastal gap, where marine fog enters the Site D valley—a proxy for the influence of fog on a locale.fMajor habitat type based on satellite imagery analysis, including mineralogy: M = mountainous; AF = alluvial fan (upper, middle, lower section of fan).gMajor habitat type based on SPI image: DP = desert pavement (surface soils mantled by gravels)—if other major habitat also available, such as a wash

or gullies, these were also noted.hMicrohabitat based on FI visible imagery: Heaves = raised soil crust, often including broken pieces; if major habitat was pebbles, then a rough estimate

of pebble density within the FI image was completed (d1 = <50% coverage in the image; d2 = �50–75%; d3 = �75%). For all habitat analyses, onlypredominant habitat type is included here.

Table 1. (continued)


6 of 16

G04S06

50–60). Conversely, within locales 10–30, FI data indicatedno chlorophyll-based life (Table 1), yet patchily distributedlichens in the rover path were readily observed in the field,albeit at low densities. The dichotomy between rover andground-based observations was a function of: (1) a smallsample size in a low abundance area; and (2) the patchydistribution of lichens at these scales (m to cm). Theseresults demonstrate that detection limits for life are afunction of rover instrumentation and sample size per unitarea—a key design and sampling parameter in ecology.[17] A similar disparity between rover versus ground-

truth observations existed for the high abundance zone inlocales 110–120 (Table 3 and Figures 3 and 4). Withinthese locales (SPSU 3), ground-truth confirmed the scienceteam’s prediction—based on IKONOS imagery and geolog-ical mapping—of a confluence region (for water) and thus apriority for biology. However, neither satellite nor ground-based rover-acquired imagery (SPI and FI) enabled thescience team to (1) predict the presence of the unique‘‘gravel bar’’ habitat (that was instantly discerned in thefield) or (2) establish the connection between the markedpatchiness in percent positive ratings (FI images) betweenthese locales and the inherent patchiness in the habitat itself(SPSU 3, Table 1). In contrast, during ground-truth it wasreadily observed that variations in percent positive ratings(moss/lichen abundance) within SPSU 3 were linked directlyto the presence/absence of gravel bar habitat: high moss/lichen abundance occurred on the gravel bars (microscaleand macroscale topographic highs covered by high pebbledensities), whereas low abundance was associated withadjacent sediment channels (topographic lows with lowpebble densities). This observed microscale to macroscalebiological patchiness and its linkages to changing ‘‘eco-hydrological’’ factors such as habitat and topography[Rodriguez-Iturbe, 2000; Lookingbill and Urban, 2004]was observed repeatedly throughout the 2005 field exper-iment at all sites.[18] Similar to SPSU 3, only in-situ ground-truth obser-

vations allowed the correlation between microbial patchi-ness in SPSU 4 and variations in microtopography and

substrate heterogeneity, among others, to be discerned.Furthermore, within SPSU 4, rover-acquired FI data showedno visible life (0% CHL) for locale 140 (located on a cm-scale topographic high), whereas ground-truth revealed thepresence of lichens a few cm away (on a cm-scale topo-graphic low). Likewise, although extreme heterogeneity inpercent positive ratings was mapped accurately by the roverwithin locales 200 to 210 (Tables 1–3 and Figures 3 and 4)—a third zone of high abundance—it took ground-truthsurveys along the rover path to understand that this micro-bial patchiness was linked to topography. More specifically,SPSU 5 low ratings were associated with a rockier flatsurface crust (relative to SPSU 6), whereas the higherabundance in SPSU 6 was associated with a small hill (alsosurface crust). Higher-scale factors may also explain theabundance shift between SPSU 5 and 6, as orbital imageryindicated SPSU 5 and 6 were within different geologicalunits (or a transition area at the boundary). No evidence forthis change in geology was evident, however, from SPI,navigation camera or FI imagery.[19] Another main habitat/abundance transition (Table 3)

mapped by the rover occurred between geological units p1and p2—from high abundance (locales 240–260a, SPSU 7)to low and patchy abundance (locales 260b–270, SPSU 8).FI data indicated locales 240–260a as highest in percentpositive ratings overall within Site D (Figure 4). This wasconfirmed by ground-truth, with virtually every surfacecovered by lichens. This zone occurred on the valley floor,which from SPI appeared to be desert pavement. However,from the FI perspective it was clearly heaved surface crusthabitat. Why this particular zone had significantly higherabundance than all others is unknown, but it may be relatedto apparent variations in materials (e.g., aeolian dust)covering the surface crust habitats that enhance microbialsurvival (e.g., moisture retention, nutrient availability). Incontrast, rover (SPI, FI) data and ground-truth clearlyshowed that the lower abundance mapped in 260b–270accompanied a shift to a rocky surface habitat perhaps lessconducive to lichen growth, although localized exceptions(tied to m-scale habitat variations) occurred.

Table 2. Biosignature Sampling Results Across Sites (Wet to Dry, From Left to Right Within Each Year)

Site

2005a 2004

F D E E^b B C

1. Total # samplesc 50 92 45 25 24 19Percent positive (simple rating scheme)d 72 50 36 20 94 952. # Visible/Chlorophyll only samplese 15 62 25 17 1 0Percent positivef 20 30 0 0 0 03. # Full samplesg 35 30 20 8 23 19Percent positiveh 94 90 80 20 96 954. Site mean (standard error) percent positive(conservative rating scheme)i

33 ± 4 30 ± 4 15 ± 3 6 ± 3 44 ± 5 47 ± 4

aSites D, E and F were completed in 2005; Sites B and C were completed in 2004.bE^: Site E excluding playa samples.cTotal number of samples where a weak or strong signal (�1 rating) in any of the 6 FI biosignature channels = positive.dPercent positive rating = total # of positive samples/total # samples X 100.eTotal number of chlorophyll and visible biosignatures (CHL) only.fPercent of CHL samples that were positive, with a 100% rating indicating both channels had strong positive signals.gTotal number of full samples, where full sample was defined as one in which all FI biosignatures were acquired. For 2004,

full sample was visible, chlorophyll, DNA and protein channels; For 2005, these 4 channels plus carbohydrate and lipid channels.hPercent of full samples that were positive, with 100% rating indicating all channels had strong positive signals.iSite mean = average of percent positive ratings for all samples at a site. This represents the most stringent/conservative measure

of a site’s relative biological abundance as mapped by the rover.


7 of 16

G04S06

[20] Site D trends in microbial abundance and distributionmapped by the rover FI mirrored those from laboratoryculture studies of heterotrophic bacteria in soils from thesame locales (Figure 5) [see also Cabrol et al., 2007b].Laboratory results indicate bacterial abundance from 10–1,000 colony-forming units/g-soil (CFU/g-soil, determinedas MPN enumerations on 1/10 strength PCA medium[Navarro-Gonzalez et al., 2003; Warren-Rhodes et al.,2007a]). Site D geometric sample mean was 15 CFU/g-soil—lower than the 2004 investigation results for Site B(32 CFU/g-soil). Similar to rover data (Figure 4), highestMPN numbers occurred at locales 210 (1 � 103 CFU/g-soil)and 110 (11 � 102 CFU/g-soil)—yet again underscoring theeffectiveness of pre-traverse selection of biological hotspotsusing satellite spectra/geological mapping and follow-the-water strategies.

3.2. Statistical Analyses and Site D Conclusions

3.2.1. Methods[21] Significant variation in microbial abundance, spatial

distribution and diversity was mapped at Site D. Themagnitude and interpretation of this heterogeneity is scale-dependent, and following the LITA field experiments, wetested various factors to explain this observed biologicalheterogeneity, using only rover-acquired datasets (Table 1):on-board percent air relative humidity (RHA); distance fromthe coastal gap (a proxy for fog influence), orbital habitattype, ground-level habitat type, and microhabitat type.[22] A mixed linear model was used (with PROC MIXED

in SAS 9.0) to evaluate whether significant differences inpercent positive ratings were observed among all samplesand SPSU, and Pearson’s correlation coefficients were alsoused to identify if a relationship existed between percent

Figure 3. BioHabitat Map overlay on Site D IKONOS VNIR false color image (1 m/pixel), whichshows habitats, geological units (yellow numbers, black boundaries), locales (white numbers), SPSU (S)and zones of high/low biological abundance mapped by the rover within Site D. Note that SPSU 1(locales 50–60, a zone of high abundance) is not shown. Within an SPSU, boxes (not to scale) indicate asingle FI sample was acquired, with green representing a positive biosignature rating (positive = �1channel had a �1 positive signal) and white a negative rating (negative = all channels have a zero rating).The gravel bar habitat is shown in green, while the main lichen fields mapped are shown in red; theboundaries of these areas are approximated from ground-truth surveys. Geological units are as follows:Fe: alluvial fan; p1—plains-forming unit (older materials), valley floor; p2—plains-forming unit(younger materials than p1) mixed with some alluvial fan materials; Fb—alluvial fan unit near coastalgap; Fa1—fan/hillslope unit near area of high sulfate concentrations in satellite spectra; Fd—alluvial fan;Fe—alluvial fan/fluvial-materials unit in confluence region postulated to have high potential for past orpresent water activity and a bottleneck for inflowing westerly marine fog through the coastal range gap(GAP). Inserted images show habitat transitions 1 and 2 marked on the Biomap. Left image set: transition1 from gravel bar habitats (SPSU 3, left image) with abundant moss and lichens to desert pavement(SPSU 4, right image) that had extremely low percent positive ratings (SPSU mean = 5%); Right imageset: habitat transition 2 from lichen field/zone of high abundance in the valley floor unit p1 (SPSU 7, leftimage) to the darker substrate (SPSU 8, right image) of p2, a zone of low and patchy abundance.


8 of 16

G04S06

positive ratings, distance to coastal gap and air relativehumidity data. All tests were adjusted for unequal samplesize and variance and statistical significance evaluated atP � 0.05.[23] As discussed earlier, total percent positive ratings

varied significantly across SPSU [F-test, df = 8, 61; F =7.83, P < 0.0001], with several zones of high and lowabundance (Figure 4). Below, linkages between these var-iations in abundance and habitat type are examined at threemain scales: (1) orbital (i.e., across geological units, >1–10sof km, satellite imagery); (2) landscape (i.e., within ageological unit, variation across and within SPSU, �30–200 m, SPI data); and (3) microscale (<10 cm - several m,i.e., variation across or within FI images).3.2.2. Orbital-Scale Analyses[24] Four zones of high abundance were mapped across

Site D, and each was associated with a particular habitattype, irrespective of their relative spatial location. Thus,from orbital analyses, microbial abundance was not ran-domly distributed. Specifically, non-alluvial fan (non-AF)units had significantly higher percent positive (46.4%)ratings than alluvial fan (AF) units (16.2%) [t-test forunequal variances, t (72) = 3.818, P = 0.0003], with threehigh abundance zones (Figures 3 and 4) associated withsurface crust type habitats. The fourth high abundance zonewas associated with the gravel bar type habitat (unit Fe). Incontrast to habitat type, no significant relationship existedbetween distance from the coastal gap (Figure 6) andpercent positive biosignature ratings [r = 0.179, p =0.1174]. This finding likely reflects the minor temporalvariation in fog hours among locales (ranging from 2–4 km from the gap).[25] Geological and mineralogical mapping (Figure 3)

accurately identified regions of biological interest andhabitat transitions a priori (as well as facilitated a posteriorianalyses). However, in several instances such mapping wasnot able to pinpoint the specific habitat type that would

predominate within a geological unit. For example, orbitalmapping clearly and correctly defined units p1 and p2 asplains-forming, and unit Fa1 as hillslope/fan in a highsulfate area (high biological priority). However, it wasunable to predict, for example, that all three units specifi-cally would be surface crust habitat, or that unit p1 wouldexhibit higher abundance than units p2 or Fa1.[26] Therefore, a systematic field-based testing of linkages

between orbital-scale geological units, ground-based habi-tats (surface and subsurface) and microbial abundanceremains integral to a catalog of predictive terrestrial analogsfor robotic biological exploration on other planets. Althoughassumptions between orbital scale geological units and life(e.g., hypersaline lakes, hot springs) have been previouslyanticipated in astrobiology research (i.e., the study of life andits distribution in the universe and one impetus for the study

Table 3. Habitats and Transitions Along the 14 km Site D Rover Traversea

Locale (SPSU) Rover Habitat Transitions and Observations

Locales 10–30 plains-forming unit p2 �1.5 km east of unit p1; mostly desert pavement,wash materials (larger rocks) and gullies;

Locales 40–60 (SPSU 1) locale 40 in transitional unit between p2 and unit M hillslopes (locales 50–60); 50–60 consist (SPI)of polygons and raised surface crusts (heaves); surface spectral results (VNIR/TIR) revealed clays with minorsulfates (see Piatek et al., submitted manuscript, 2007); first habitat transition observed (SPI, FI); ground-truthconfirmed this transition as biologically significant;

Locales 90–100 (SPSU 2) DP mixed with raised surface materials (AF unit Fe); dark rocks oxides and sulfates (VNIR); second habitattransition observed;

Locales 110–120 (SPSU 3) within geological units Fb (AF) and Fe (AF-fluvial materials) in priority region (‘‘confluence’’ area andfog ‘‘bottleneck’’) from geological mapping; SPI shows DP (clays, VNIR/TIR) with many gullies; FI imagesindicated high pebble density; third habitat transition observed; Ground truth revealed unique ‘‘gravel bar’’ habitat;

Locales 121–140 (SPSU 4) DP (AF unit Fb) habitats. FI shows habitat transition at locale 121 that separated 110–120 from 130–140 (SPSU 4),with more exposed crust;

Locales 190–200 (SPSU 5) mixed AF/younger plains-forming materials in geological unit p2;Locales 200–210 (SPSU 6) geological unit Fa1 (‘‘sulfate hill’’) identified from satellite mineralogy as a priority due to potential sulfates signatures;

SPI and FI data showed hillslope habitat with polygons and heaves (e.g., locales 40–60); FI also showed raisedporous-looking surface crust with few rocks. An important habitat transition from DP in previous locales;

Locales 240–260 (SPSU 7) traversed from AF/hillslope unit Fa1 to p1, but SPI and FI data contained similar habitat as locales 190–210(although there were some microhabitat differences);

Locales 260–270 (SPSU 8) shift from older valley floor to p2 (Figure 2), with habitat transition evident in SPI (exposed crust with tiny pebbles)and FI; this region was hypothesized as a high priority for fog embankment;

Locales 290–300 (SPSU 9) DP habitat straddling alluvial fan units Fb, Fd and Fe.aDP = desert pavement; AF = alluvial fan. Locales in italics indicate zones of high microbial abundance. Unless otherwise noted, mineralogy derived

from ground-based VNIR/TIR.

Figure 4. Variation in Site D SPSU mean (standard error)percent positive biosignature ratings. Locale numbers areshown in black and corresponding habitat in grey.


9 of 16

G04S06

of terrestrial analogs), to our knowledge this is the firstorbital-scale testing of these associations.3.2.3. SPI-Based Analyses[27] Heterogeneity in percent positive biosignature rat-

ings was also mapped by the rover at smaller scales, such asthose within the gravel bar habitat (SPSU 3), where FIsample ratings ranged from 0% to 100% in response to m tocm-scale geological transitions. Similar shifts were notedwithin SPSU 8, where FI ratings also changed with a shift inhabitat. Such rating variations were not fully understooduntil after ground-truth observations, which revealed higherabundance was associated with a habitat transition (m-scale)to gypsum heaved crusts.[28] These observations for SPSU 3 and 8, among many

others, confirm that non-random microbial distributions alsoexist at landscape/SPI scales. Such non-random spatialdistributions are partially explained by ecohydrologicalfactors such as climate, topography (e.g., affecting wateravailability) and geology (e.g., substrate changes). Forexample, during the 2004 field experiment at Site B, roverdata clearly tied lichen populations to areas with fog[Warren-Rhodes et al., 2007a]. This connection was furtherconfirmed in-situ for Site D using rover climate data, whichindicated that locales where fog was present (as measuredby rover RHA � 95%, liquid water condensation on rover

leaf wetness sensors, and/or gaps in solar insolation) sup-ported robust lichen communities. Post-operations analysis forSite D found a significant correlation between rover RHA (andthus also for co-varying solar insolation and leaf wetness data)and SPSU percent positive ratings [r = �0.295, p = 0.0494].[29] While some insights into the linkages between eco-

hydrology and life were deduced from ground-based cli-mate data and SPI imagery, a much clearer understanding ofmicrobial spatial heterogeneity at landscape scales wasobtained during post-ops ground-truth. Surprisingly, SPIimagery habitat analysis (�1 m to 200 m2 field of view)was less predictive of biologically rich locales than antici-pated. Through the lens of the SPI camera, which providedthe most ‘‘field-oriented’’ view of the traverse, landscape-based habitat analysis concluded that 6 of the 9 SPSU inSite D were desert pavement, thus providing data onhabitats comprising the bulk of surveyed area and importantcontext along the traverse. However, beyond this data, SPIhabitat analysis was ineffective in predicting a priorior explaining a posteriori traverse-level microbial spatialdistributions. Instead, for example, only ground-truth eluci-dated differences accounting for ratings extremes betweenSPSU 5 (meanSPSU5 = 6.6 ± 12.7% rating) and SPSU 6(meanSPSU6 = 60.8 ± 11.6%) that accompanied a terrainchange from surface crust in a flat area to that on a smallhill. This macrotopography change (potentially enhancingfog deposition) was not easily discerned, or was simplymissed, during the rover traverse.[30] SPI-scale habitat types also could not be linked

statistically to areas of higher microbial abundance, possiblyas a result of insufficient habitat type resolution. Notwith-standing this issue, it was apparent during ground-truth thatSPI imagery simply does not capture or relay to scientists ina sufficiently rapid manner (e.g., 3-D, slope changes)contextual clues about habitats and environment that asimple 5-minute walk along a traverse does. Embeddingsuch information into SPI imagery might be a first steptowards resolution of this disparity. Ground-truth observa-tions, in contrast, readily revealed that landscape variables,such as a change in topography, were correlated withmicrobial spatial patterns—not surprising given previous

Table 4. Least Squares Means Differences in SPSU, Site Da

SPSU (Transect) Mean Pairwise Comparisons N

2 5.25 A**** 94 5.80 A**** 105 6.60 AB*** 69 10.80 AB*** 108 34.29 ABCD* 71 34.38 *BCD* 83 48.88 **CD* 86 60.83 ***DE 67 86.00 ****E 8

aN = sample number. Note: Means with the same letters are notsignificantly different from each other.

Figure 5. Regression results for Site D locales for rover FIpercent positive biosignature ratings versus laboratorycultures of soil heterotrophic bacteria reported as colony-forming units per gram soil (CFU/g-soil). For laboratoryculture methods, see Warren-Rhodes et al. [2007a].

Figure 6. Mean (standard error) percent positive biosigna-ture ratings for SPSU versus distance from coastal gap,Site D. Locales shown in black and SPSU in grey.


10 of 16

G04S06

ecohydrology research [Rodriguez-Iturbe, 2000; Guswa etal., 2002; Lookingbill and Urban, 2004]. Likewise, land-scape variables and their spatial heterogeneity explained aposteriori many changes at the FI scale. Yet, the connectionsbetween the two scales were often either missing or missedduring remote science operations.3.2.4. Microscale (FI) Analyses[31] LITA 2004 results (Sites B and C) indicated a

predictive correlation between orbital and microscale hab-itats and transitions [Warren-Rhodes et al., 2007a]. Site D2005 rover data further confirmed this insight. Similar toorbital findings, FI analyses showed raised crust micro-habitats had significantly higher total percent positiveratings than pebble microenvironments [F-test, df = 2, 74;F = 10.32, P < 0.0001]—thus linking the two separateorbital and FI datasets. In contrast, no differences existedbetween pebble microhabitats of varying densities andpercent positive ratings [test for unequal variances, t (74) =�1.78, P = 0.079]. This may be a function of analyticalresolution, since microhabitat types (Table 1) were derivedfrom simple classification of FI images based on rapidassessments (e.g., high vs. low pebble density) duringremote operations and post-ops. In contrast, detailed sedi-mentology analyses (conducted on a limited number ofsamples, Figure 7) may yield greater insights into biotic-microhabitat relationships. Lastly, within single FI images(cm to mm scales), the strength of positive CHL and DPLCsignals was also often non-uniform, although the degree

differed amongst biosignatures (see Weinstein et al., sub-mitted manuscript, 2007).

3.3. Across-Site Comparisons: Site D (Coastal), Site E(Hyperarid Core) and Site F (Interior)

[32] In contrast to coastal Site D, Site E was situatedwithin the hyperarid core of the Atacama’s central interiordepression (Figure 1). These extremely dry conditions werereflected in Site E’s overall lower biosignature rating (36%of 45 samples were positive, Table 2), which was the lowestamong sites. Only a brief overview of Site E (and F) ispresented below to provide focus to across-site habitat andbiological comparisons.3.3.1. Site E Results and Conclusions[33] Two main regions of biological interest were identi-

fied (pre-traverse) within Site E (Figure 8): (1) the northerntopographic low (Pvb plains-forming unit of brightalbedo)—hypothesized to be a salar; and (2) the southernassemblage of drainages (confluence), including the Pmb(plains-forming, medium albedo) unit. These regions wereselected primarily based on evidence for past or presentaqueous activity (‘follow-the-water’ strategy), includingpossible influence by fog, and promising orbital mineralogy,such as areas of relatively high sulfates and quartz concen-trations (Piatek et al., submitted manuscript, 2007). Nochlorophyll signatures were detected for Site C in orbitalsatellite spectra.[34] Table 5 lists the three main habitats mapped within

Site E: (1) alluvial fans consisting of desert pavement (SPI

Figure 7. Microscale sedimentology of rover-acquired FI images was assessed in a process analogousto that of thin section analysis by reflected light microscopy [Butler et al., 2001; Kanduri et al., 2005].For the Atacama soils, each frame was processed, representative areas were selected, and the longest/shortest axes of individual grains (or other fabric elements) were measured to create histograms andcumulative frequency distributions [Cabrol et al., 2007a]. Size fractions were distinguished using theWentworth grain size scale [Wentworth, 1922; Herkenhoff et al., 2002; Cabrol et al., 2006]. Resultsshowed distinct differences for locale 100 (right), which contained no positive biosignatures for life (0%rating) versus locale 110 (left), where strong positive signals (100% rating) were detected, corresponding tothe presence of abundant moss. Grain size count includes pebbles (2 mm–64 mm), sand (62 mm–2 mm),silt (4 mm–62 mm) and clay (1 mm–4 mm).


11 of 16

G04S06

and FI scales); (2) desert pavement in the pwS2 geologicunit (wind-streaked plains-forming materials); and (3) playaof the Pvb (non-AF) unit (SPI and FI data (Weinstein et al.,submitted manuscript, 2007)). Strikingly, no positive chlo-rophyll signatures were mapped in the 19 km traverse,irrespective of habitat. Ground-truth reinforced this obser-vation, with Site E the only site where no chlorophyll wasdetected by the rover sensors.[35] Similar to Site D, percent positive ratings (DPLC) for

non-AF habitats (e.g., playa) were markedly higher than forAF/desert pavement habitat, further confirming the ability topredict zones of high relative abundance from orbital habitatanalyses. Indeed, the playa was the main biotic repositorywithin Site E, with the site’s mean percent positive rating(6 ± 3%, Table 2) �5 times lower than Site D or F.However, no significant differences in percent positiveratings across comparable desert pavement habitats werefound between Site E and D (Table 5). As at Site D, smaller-scale patchiness (within a geological unit and/or SPSU) was

also mapped at Site E, with lower biosignature ratings insouthern (SPSU 3 = mean DPLC rating 63 ± 19%)versus northern playa samples (SPSU 4, mean DPLC rating76 ± 8%).[36] Unexpectedly, FI data showed no evidence for

lichens or cyanobacteria within dry playa sediments andplowed surfaces (�1 cm deep). Post-ops ground-truthlikewise detected no photosynthetic communities, eventhough subsurface sediments were moist to several cm.However, biologists present at the playa following unusualwinter rains (after science operations had concluded) ob-served green-tinged layers near the surface, indicatingpossible ephemeral or mobile [Pringault and Garcia-Pichel,2004] algal and/or cyanobacterial populations. Post-opsground-truth south of the rover traverse (Figure 8, unit M)also revealed localized lichen populations and evaporites(halites) in the Pmb unit, but these habitats were not reachedduring science operations.

Figure 8. ASTER VNIR false color image (15 m/pixel) of Site E Regions of Interest Map. Geologicunits include the Pvb (plains-forming) unit of bright albedo—hypothesized to be a salar or other type ofsedimentary feature—the pwS wind-streaked plains-forming materials) unit, and the Pmb (plains-forming) unit. Other units mapped include: M = mountainous, D = dissected, F = alluvial fan, and peS =etched unit. These regions were targets of interest based on past or present aqueous activity (‘follow-the-water,’ light and dark blue dots), including possible influence by fog (hypothesized flow shown in bluearrows). Areas of interest targeted from orbital mineralogical spectra are shown in green (orbitalcomposition results were not conclusive) and red (sulfates, quartz, volcanics). No chlorophyll signatureswere detected in the satellite spectra. The orange dot represents the rover’s initial estimated position onthe first day of operations.


12 of 16

G04S06

[37] On-board rover climate data did not indicate strongvariations in climate (max./min. temperature and RHA,Table 6) between Sites E and D. Site E leaf wetness, solarinsolation and RHA data indicated the likely presence of fog(Figure 9), which was confirmed (post-ops) by engineeringteams who observed the influx of nightly marine fog aspredicted from orbital analyses (Figure 8). During theseevents, fog was dense at unit M but of lower intensity andfrequency near the Pvb unit, which likely explains lichenpresence in unit M but absence in the rover path.[38] Similar to Site D, microbial life at Site E was patchy

and non-random, often being associated with substrates thatefficiently retain scarce fog moisture and/or intermittentrainfall. Site E results also support the rationale for asweeping reconnaissance of important units of biologicalinterest prior to intensive sampling. This strategy is partic-ularly important for extremely dry environments, wheresurface and/or photosynthetic microbes are often rare and/or localized [Warren-Rhodes et al., 2006; Wierzchos et al.,2006]. For such sites, maximizing the number of biosigna-tures tested is likewise necessary to detect microbial pop-ulations: in only three of 20 Site E samples were all fournon-chlorophyll channels strongly positive. Thus, a major-ity of positive ratings would have been missed had only 1biosignature been used (as a SOTF trigger). Lastly, Site Eresults demonstrate the importance of temporal and spatialsampling, i.e., when you sample (e.g., after a rainfall event,during ‘wet season’) may be as important as where yousample.3.3.2. Site F Results[39] Similar to Site E, Site F is located in the Atacama

extreme dry interior (Figure 1), but rover mapping revealeda stark difference in percent positive ratings (72% of 50samples were positive, Table 2). Satellite imagery showedSite F to be highly diverse geologically, with hilly terrain,

massive drainage channels, alluvial fans and numerous highalbedo units suggestive of past/present hydrologic activity(playas, salars). Satellite spectra also indicated high miner-alogical diversity, with volcanics, quartz and sulfates (i.e.,evaporite materials such as gypsum) signatures detected—the latter a high priority for biology [Warren-Rhodes et al.,2007a; Piatek et al., submitted manuscript, 2007].[40] Another unique Site F feature was its high elevation

(min. 1700 m). The science team hypothesized Site F’scombined altitude, distance from the coast and diversemineralogy would impact both climate and habitats in thefollowing manner: (1) marine fog would be negligibleand preclude lichens; (2) habitat diversity would increaseover previous sites (including possible hot springs);and (3) biological diversity would increase (e.g., higherplants, desert crusts), particularly in habitats supported bywinter snow/rain and potential runoff/groundwater from theAndes. Given its proximity to Site C, RHA was forecast tobe low during Site F operations (spring in Chile). Asdescribed below, many of these hypotheses were confirmed,highlighting the increased predictive power of a scienceteam over time as knowledge of ground-based habitats,climate and biology is strengthened.[41] In contrast to the habitat diversity suggested by

orbital geology and mineralogy, and despite the distance(56 km) covered, the Site F traverse mapped only a singletype of habitat: desert pavement. This homogeneity providedan opportunity to illustrate the rover’s ability to detect theeffects of greater water availability, with desert pavementhabitats in this water-rich region supporting rich biologicalcommunities and having higher percent positive ratingseven than Site E playa or certain Site D surface crusthabitats (Table 5). Within Site F, pavement habitats east ofa hilly unit and within a massive drainage area (locales730–910) had markedly higher plant and microbial abun-dance than comparable habitats to the west (locales 1150–1250) (Table 5) (see Weinstein et al., submitted manuscript,2007). We hypothesized prior to the traverse that such abiological divide would exist due to a rainshadow effect of

Table 5. Across-Site (D, Wet Coastal; E, Dry Interior; F, Wet

Interior) Habitat Comparisons, Including Both Simple (Total

Percent +) Versus More Conservative (Mean Total Percent +)

Ratingsa

LocalesTotal% +

Mean Total% + Major Habitat Type

D 250–260 100 86 Surface crust/valley floorF 840–880 100 64 DP (confluence, east, plants)D 200–210 83 61 Surface crust/hillslopeD 110–120 75 50 Confluence/gravel barE 670–680 55 47 Playa (north)F 730–790 100 45 DP (east)E 640–680 63 44 Playa (all samples)D 50–60 63 34 Surface crust/hillslopeD 260–270 57 34 DP/valley floorF 890–910 67 33 DP (east, plants)F 1150–1170 71 29 DP, westF 800–810 63 24 DP (east)D 290–300 20 11 DPD 130–140 20 6 DPE 350–360 10 7 DPE 330–340 16 2 DP

aDP = desert pavement (surface soils mantled by gravels). Total percentpositive is calculated as # positive samples within habitat/total # sampleswithin habitat X 100. Mean total percent positive is calculated as theaverage of sample ratings for an SPSU or given locales within a singlehabitat type.

Table 6. Across-Site Comparisons of Climate and Biologya

B C D E F

Rover-acquired dataLichens detected during traverse? Y N Y Nb NMaximum RHA (%) �95 25 �95 �95 15Minimum RHA (%) 70 3 5 1 3Maximum Air Temperature (oC) 26 23 40 36 25Minimum Air Temperature (oC) 10 4 0 3 4Maximum Wind speed (m/s) 6 16 6 13 25Presence of Fog (leaf wetness sensor) na na Y Y NFog (confirmed in SPI image) Y N Y Y N

Ground-truth/historical dataMAP (mm/yr) 2 5 2 2 >5Distance to Pacific Coastal range (km) <5 80 <5 50 100Distance to Andes (km) 170 110 155 150 95aHere na = data not available; Sites B, D = ‘wet’ (fog) coastal; Site C =

dry interior; Site E = extremely dry interior; Site F = wet (relative to Sites Cand E) interior. RHA = air relative humidity from on-board sensor, with�95% indicating presence of liquid water (usually fog, as confirmed bySPI, leaf wetness sensor and solar insolation data). Mean annualprecipitation (MAP) from long-term historical climate data from DireccionMeteorologica de Chile.

bLichens observed during ground truth at Site E outside of traverse area.


13 of 16

G04S06

the hills, with eastern habitats (2300–2500 m altitude)receiving more precipitation during winter from the Andes(snow was visible in SPI images). Rover-acquired FI, SPIand navigation camera images, along with post-ops ground-truth, confirmed this climate effect: eastern habitats hadabundant plant life, including fields of wildflowers, whereasfew plants were present in the west.[42] Table 5 also shows significant smaller-scale patchi-

ness amongst Site F eastern desert pavement habitats(i.e., across and within SPSU) in relation to varying wateravailability (e.g., higher CHL in/near drainage channels,where most plants were concentrated). Navigation cameramovies highlighted these water-biota linkages and providedimportant environmental context for sampling results—critical given the large distances between SPSU (i.e., 3–5 km). We assume that other factors controlling abundance,as described for Site D (e.g., macrotopography/microtopog-raphy), likely explain patchiness at Site F.[43] Overall, Site F biosignature ratings reflect a more

biologically rich environment than Sites D or E, explainedin part, as predicted, by the presence of higher plants.However, even when SPSU with plants are excluded, SiteF percent positive ratings are higher than Site D or E acrosssimilar habitats (e.g., Table 5). As predicted given the lackof fog, lichens were not detected (see below). It should benoted that chasmolithic cyanobacteria (locale 701) locatedinside calcite rocks were missed during the traverse. Theseorganisms were located (post-ops) outside the FI samplingarea and found only after an extensive search, but the lackof detection of these communities highlights the difficultythat rover-based searches that are limited both in the numberof samples that can be processed and/or in samplingtechniques (e.g., inability to mimic a field biologist employ-ing a rock hammer) will encounter in detecting microbiallife that utilize avoidance strategies to survive extremeconditions.

3.3.3. Site Comparisons[44] Several biosignature rating comparisons of Sites D, E

and F are possible from rover FI data—from the simplestand least conservative measure (i.e., number of positives,where positive = any rating �1, any category) to morespecific ratings (e.g., mean total percent positive). Regard-less of the measure, overall trends between sites and acrosshabitats were similar (Table 5): (1) the relative abundanceand diversity of life dropped from wet to dry sites, espe-cially due to the loss of lichens, moss and higher plants; and(2) desert pavement/AF habitats generally had lower abun-dance (except in wet environments, e.g., Site F drainageseast) than non-AF habitats (e.g., heaved surface crusts,playa).[45] Across-site comparisons in the Atacama are readily

understood through coastal versus interior desert moistureavailability. Based solely on rover data, lichen distributionis clearly tied to the reach of coastal fog, except for Site E,where fog was recorded but lichens were observed onlyoutside the traverse area (Table 6, SPI and RHA, solarinsolation and leaf wetness). Although no apparent connec-tion existed between other rover-acquired climate data andtotal percent positive ratings of the five sites (Tables 2 and6), post-operations analyses of Atacama Desert historicalclimate data readily explain the biotic abundance anddistribution trends mapped by the rover. As Figure 1illustrates, biotic trends follow the longitudinal west-to-east‘wet-dry-wet’ physio-geographical and climate transi-tions—from fog-rich coastal zones (highest abundance,lichens/moss) in the west (Sites B and D), to hyperaridcentral deserts (lowest abundance, Site E) to higher-altitudeeastern Andes (closer to the Pre-cordillera) dry deserts(Site F), where winter precipitation (and likely groundwater)supports limited plant growth and higher microbial abun-dance relative to the most arid core [Miller, 1976; Rundel etal., 1991; Larrain et al., 2002; Latorre, 2002; Latorre et al.,2003; Houston, 2006].

4. Lessons Learned

[46] The LITA project holds significant lessons for themapping of terrestrial extreme environments and the searchfor and detection of life on Mars [see also Cabrol et al.,2007b]. While rover-derived biological trends matchedground-truth and previous field-based studies in the AtacamaDesert [Latorre et al., 2002; Navarro-Gonzalez et al.,2003], LITA results suggest that microbial abundance inportions of the Atacama may be more varied and less rarethan previously anticipated [see also Lester et al., 2007].This result was unexpected, but may be partially explainedby differences in microbial detection methods (e.g., culturevs. culture-independent vs. in-situ fluorescence detection).[47] Secondly, microbial abundance and spatial distribu-

tion were patchy at all scales investigated, with intra- andinter-site differences partly explained by environmental,geological, mineralogical and other factors (e.g., topogra-phy). The importance of these ecohydrological variables increating non-random microbial patterns is not surprisingbased on landscape ecology studies of desert vegetation[Rodriguez-Iturbe et al., 1999; Rodriguez-Iturbe, 2000;Lookingbill and Urban, 2004; Ludgwig et al., 2005].

Figure 9. On-board rover environmental sensors forsol 10, Site E. Air relative humidity (thick black line,RHA � 95%), solar insolation (thin black line, brokencurve) and leaf wetness (thin grey line, Volts � 0.2) dataindicate the presence of, and liquid water condensationfrom, fog from �midnight to 6:00 am at Site E. The thickgrey line denotes air temperature (TA, �C).


14 of 16

G04S06

[48] As the LITA 2005 study demonstrates, at orbital tomicrobial scales, diversity and abundance are inextricablylinked to broad habitat types, with alluvial fan/desertpavement habitats generally having lower abundance thannon-AF units. Significantly, many of these non-AF units areshaped by past/present water activity and identifiable fromorbital imagery and mineralogy, including playa, hypersa-line deposits and surface crust habitats. To readers familiarwith astrobiology, this result supports targeting ‘‘analog’’environments through satellite reconnaissance, geologicmapping and a ‘follow-the-water’ strategy to search for lifeon other planets. (The relevance of LITA explorationstrategies for subsurface life on Mars (if any) is discussedby Cabrol et al. [2007b].)[49] While orbital analyses showed some limitation, geo-

logical and mineralogical satellite mapping results wereaccurately reflected in microscale (FI-based) habitat com-positions (e.g., Table 1), trends and transitions, makingthem important guides for ground-based sampling. However,their predictive value in homogeneous environments (e.g.,km-scale pavement in deserts), and particularly at smallerlandscape scales, remains uncertain, owing to the complexityof underlying causes creating biological patchiness indeserts.[50] Overall, the LITA experiment clearly shows that an

exploration strategy based on long-distance surveys ofregions of interest (selected from orbit) is a highly effectiveapproach to the search for life across vast geographical scales.One critical issue raised by this approach is the balancebetween reconnaissance, which favors rapid sampling tocover distance, and biological sampling, which demandssignificant time and effort. Optimizing this ‘‘science/mobilitytradeoff’’ [Cabrol et al., 2007b] requires (1) a clear and well-designed sampling plan; (2) pre-traverse refinement andtesting of sampling methods (i.e., optimizing sample num-ber, distance between samples, etc.); and (3) applicationof the chosen ground-based sampling method (e.g., SPSU) ina systematic, consistent, comprehensive and strategic way toallow standardized comparisons andmaximize scientific gainrelative to efficiency.[51] Conversely, LITA mapping results also highlight the

need for intensive sampling within locale/geological unitareas (e.g., 500 m2) prior to moving to a new region. Theutility of this approach is highlighted in the 2005 results(Table 5), when more intensive SPSU sampling (every 30 m,n = 7 samples per SPSU) was instituted compared to 2004(every 500m, n = 1–2 samples). This single sampling changefundamentally transformed the quality of science dataacquired during reconnaissance. Consequently, (1) 2005rover sampling maps were highly accurate at geologicalunit and individual locale (�200 m2) scales (confirmed byground-truth), whereas biota was not fully characterized in2004; and (2) percent positive ratings in 2005 systematicallyquantified and distinguished between wet and dry sites,which was not shown during the previous year.[52] The above findings highlight the need for a large

sample size to detect and characterize microbial populationsand hotspots across multiple scales. Furthermore, theysuggest a comprehensive comparative study of optimalsample size, number, spacing, replication, and competingmicrobial detection tools to map life in extreme environ-ments would yield substantial benefits. Although such

studies and techniques—standard in landscape ecology—are only recently being applied to microbial ecology, theirutility cannot be understated. Robotic exploration teamsstand to benefit greatly from these methods, particularlythose relating to the sampling of rare and endangeredspecies [Thompson, 2004]. Such optimization studies couldvastly improve the search for life in extreme deserts—whether on Earth or Mars—where microbial populationsare likely to be sparse and/or spatially patchy [Cockell andStokes, 2004; Warren-Rhodes et al., 2007b].[53] The fundamental lesson the LITA experiment teaches

is that the spatial distribution of microbial communitieswithin the driest desert environments on Earth is exceed-ingly complex at the local scale. Predicting where and whymicrobial oases exist remains difficult. However, as thestudy of extreme deserts progresses, microbial life begins toappear less rare and less random than anticipated. Althoughdiscerning microbial patterns is challenging, it is oftennonetheless inherently and ultimately predictable. Under-standing and utilizing this predictability to hone planetaryexploration strategies and optimize the search for life onMars should comprise a major science and engineeringthrust for future astrobiology research.

[54] Acknowledgments. The authors thank the LITA rover engineer-ing team for their unstinting efforts in the field. We also acknowledgeinvaluable communication assistance by the project’s IT Eventscope team.This project was supported by NASA grants SOTF NNG04-GB66G andLITA NAG5-12890.

ReferencesBegon, M., J. Harper, and C. Townsend (1990), Ecology: Individuals,Populations, and Communities, Blackwell Sci., London.

Butler, J. B., S. N. Lane, and J. H. Chandler (2001), Automated extractionof grain-size data from gravel surfaces using digital image processing forhydraulic research, J. Hydraul. Res, 39(1), 1–11.

Cabrol, N. A., J. D. Farmer, E. A. Grin, L. Richter, L. Soderblom, R. Li,K. Herkenhoff, G. A. Landis, and R. E. Arvidson (2006), Aqueousprocesses at Gusev Crater inferred from physical properties of rocks andsoils along the Spirit traverse, J. Geophys. Res., 111, E02S20, doi:10.1029/2005JE002490.

Cabrol, N. A., et al. (2007a), Soil sedimentology, textures and dynamics atGusev Crater from Spirit’s microscopic imager, Proc. Lunar Planet. Sci.Conf., XXXVIII, Abstract 1784.

Cabrol, N. A., et al. (2007b), Life in the Atacama: Searching for life withrovers (science overview), J. Geophys. Res., 112, G04S02, doi:10.1029/2006JG000298.

Cockell, C., and D. Stokes (2004), Widespread colonization by polar hy-poliths, Nature, 431, 414.

Dale, M. (1999), Spatial Pattern Analysis in Plant Ecology, CambridgeUniv. Press, New York.

Green, J., and B. Bohannan (2006), Spatial scaling of microbial diversity,Trends Ecol. Evol, 21, 501–507.

Grundmann, G. (2004), Spatial scales of soil bacterial diversity—the size ofa clone, FEMS Microbiol. Ecol., 48, 119–127.

Guswa, A., M. Celia, and I. Rodriguez-Iturbe (2002), Models of soil moist-ure dynamics in ecohydrology: A comparative study, Water Resour. Res.,38(9), 1166, doi:10.1029/2001WR000826.

Herkenhoff, K. E., et al. (2002), Textures of the soils and rocks at Gusevcrater from Spirit’s Microscopic Imager, Science, 305, 824–825.

Houston, J. (2006), Variability in precipitation in the Atacama Desert: Itscauses and hydrological impact, Int. J. Climatol., 26, 2181–2198.

Kanduri, A. K., G. Thomas, N. A. Cabrol, E. A. Grin, and R. C. Anderson(2005), The (In) accuracy of novice rover operations’ perception of ob-stacle height from microscopic images, SMCA04-08-0222, 1–8.

Krebs, C. (1998), Ecological Methodology, 2nd ed., Addison-Wesley,Boston, Mass.

Larrain, H., H. Velasquez, P. Cereceda, R. Espejo, R. Pinto, P. Osses, andR. S. Schemenauer (2002), Fog measurements at the site ‘‘Falda Verde’’north of Chanaral with other fog stations of Chile, Atmos. Res., 64,273–284.


15 of 16

G04S06

Latorre, C. (2002), Clima y Vegetacion del Desierto de Atacama Durante elCuaternario Tardıo, II Region, Chile, Ph.D. dissertation, Univ. de Chile,Santiago.

Latorre, C., J. Betancourt, K. Rylander, and J. Quade (2002), Vegetationinvasions into absolute desert: a 45000 yr rodent middden record from theCalama-Salar de Atacama basins, northern Chile (lat 22�–24�S), Geol.Soc. Am. Bull., 114, 266–349.

Latorre, C., J. Betancourt, K. Rylander, J. Quade, and O. Matthei (2003), Avegetation history from the arid prepuna of northern Chile (22–23�S)over the last 13,500 years, Palaeogeogr. Palaeoclimatol. Palaeoecol.,194, 223–246.

Lester, E., M. Satomi, and A. Ponce (2007), Microflora of extreme aridAtacama Desert soils, Soil Biol. Biochem., 39, 704–708.

Lookingbill, T., and D. Urban (2004), An empirical approach towards im-proved spatial estimates of soil moisture for vegetation analysis, Land-scape Ecol., 19, 417–433.

Ludgwig, J., B. Wilcox, D. Breshears, D. Tongway, and A. Imeson (2005),Vegetation patches and runoff-erosion as interacting eco-hydrologicalprocesses in semi-arid landscapes, Ecology, 86, 288–297.

Miller, A. (1976), The Climate of Chile, in World Survey of Climatology,vol. 12,Climates of Central and South America, edited by R. Schwerdtfeger,pp. 113–145, Elsevier, Amsterdam.

Navarro-Gonzalez, R., et al. (2003), Mars-like soils in the Atacama Desert,Chile, and the dry limit of microbial life, Science, 302, 933–1096.

Nunan, N., K. Wu, I. Young, J. Crawford, and K. Ritz (2002), In situ spatialpatterns of soil bacterial populations, mapped at multiple scales, in anarable soil, Microbial Ecol., 44, 296–305.

Nunan, N., K. Wu, I. Young, J. Crawford, and K. Ritz (2003), Spatialdistribution of bacterial communities and their relationships with themicro-architecture of soil, FEMS Microbiol. Ecol., 44, 203–215.

Pringault, O., and F. Garcia-Pichel (2004), Hydrotaxis of cyanobacteria indesert crusts, Microbial Ecol., 47, 366–373.

Riquelme, R., J. Martinod, G. Herail, J. Darrozes, and R. Charrier (2003), Ageomorphological approach to determining the Neogen to recent tectonicdeformation in the coastal cordillera of northern Chile (Atacama), Tecto-nophysics, 361, 255–275.

Rodriguez-Iturbe, I. (2000), Ecohydrology: A hydrological perspective ofclimate-soil-vegetation dynamics, Water Resour. Res., 36, 3–10.

Rodriguez-Iturbe, I., P. D’Odorico, A. Porporato, and L. Ridolfi (1999), Onthe spatial and temporal links between vegetation, climate and soil moist-ure, Water Resour. Res., 35, 3709–3722.

Rundel, P., M. Dillon, B. Palma, H. Mooney, S. Gulman, and J. Ehleringer(1991), The phytogeography and ecology of the coastal Atacama andPeruvian deserts, Aliso, 13, 1–49.

Thompson, W. (2004), Sampling Rare or Elusive Species: Concepts, De-signs and Techniques for Estimating Population Parameters, IslandPress, London.

Turner, M., R. Gardner, and R. O’Neill (2001), Landscape Ecology inTheory and Practice, Springer-Verlag, New York.

Warren-Rhodes, K. A., K. Rhodes, S. Pointing, S. Ewing, D. Lacap,B. Gomez-Silva, R. Amundson, E. I. Friedmann, and C. P. McKay (2006),Hypolithic cyanobacteria, dry limit of photosynthesis and microbialecology in the hyperarid Atacama Desert, Microb. Ecol, 52, 389–398.

Warren-Rhodes, K., et al. (2007a), Searching for microbial life remotely:Satellite-to-rover habitat mapping in the Atacama Desert, Chile, J. Geo-phys. Res., 112, G04S05, doi:10.1029/2006JG000283.

Warren-Rhodes, K., et al. (2007b), Ecology and spatial pattern of cyano-bacterial community island-patches in the Atacama Desert, Chile,J. Geophys. Res., doi:10.1029/2006JG000305, in press.

Wentworth, C. K. (1922), A scale of grade and class terms for clasticsediments, J. Geol., 30, 377–392.

Wierzchos, J., C. Ascaso, and C. P. McKay (2006), Endolithic cyanobac-teria in halite rocks from the hyperarid core of the Atacama Desert,Astrobiology, 6(3), 1–8.

��L. N. Boyle and G. Thomas, Department of Mechanical and Industrial

Engineering, University of Iowa, Iowa City, IA 52242, USA.N. A. Cabrol and E. Grin, SETI Institute, Mountain View, CA 94043,

USA.P. Coppin, Eventscope, CarnegieMellonUniversity, Pittsburgh, PA 15213,

USA.C. Diaz, Universidad Catolica del Norte, 0610 Antofagasta, Chile.J. Dohm and M. Wyatt, Department of Hydrology and Water Resources,

University of Arizona, Tucson, AZ 85721, USA.S. Emani, L. A. Ernst, G. Fisher, E. Minkley, D. Pane, A. S. Waggoner,

and S. Weinstein, Molecular Biosensor and Imaging Center, CarnegieMellon University, Pittsburgh, PA 15213, USA.A. Hock, Department of Earth and Space Sciences, University of

California, Los Angeles, CA 90095, USA.J. Moersch and J. L. Piatek, Department of Earth and Planetary Sciences,

University of Tennessee, Knoxville, TN 37996, USA.G. G. Oril, T. Smith, K. Stubbs, M. Wagner, and D. S. Wettergreen,

Robotics Institute, Carnegie Mellon University, Pittsburgh, PA 15213,USA.K. Warren-Rhodes, NASA Ames Research Center, MS 245-3, Moffett

Field, CA 94035, USA. ([email protected])


16 of 16

G04S06

robotic ecological mapping: habitats and the search for life in the atacama desert

Documents