Monitoring Spatiotemporal Surface Soil Moisture Variations During Dry Seasons in Central America With Multisensor Cascade Data Fusion

of 16
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Entertainment & Media

Published:

Views: 0 | Pages: 16

Extension: PDF | Download: 0

Share
Description
—Soil moisture is a critical element in the hydrological cycle, which is intimately tied to agriculture production, ecosystem integrity, and hydrological cycle. Point measurements of soil moisture samples are laborious, costly, and inefficient.
Tags
Transcript
  4340 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 11, NOVEMBER 2014 Monitoring Spatiotemporal Surface Soil MoistureVariations During Dry Seasons in Central AmericaWith Multisensor Cascade Data Fusion Chi-Farn Chen, Miguel Conrado Valdez, Ni-Bin Chang, Li-Yu Chang, and Pei-Yao Yuan  Abstract— Soil moisture is a critical element in the hydrologicalcycle, which is intimately tied to agriculture production, ecosys-tem integrity, and hydrological cycle. Point measurements of soilmoisture samples are laborious, costly, and inefficient. Remotesensing technologies are capable of conducting soil moisture map-ping at the regional scale. The advanced microwave scanningradiometer on earth observing system (AMSR-E) provides globalsurface soil moisture (SSM) products with the spatial resolutionof 25 km which is not sufficient enough to meet the demandfor various local-scale applications. This study refines AMSR-ESSM data with normalized multiband drought index (NMDI)derived from the moderate resolution imaging spectroradiometer(MODIS) data to provide fused SSM product with finer spatialresolution that can be up to 1 km. Practical implementation of thisdata fusion method was carried out in Central America Isthmusregion to generate the SSM maps with the spatial resolution of 1 km during the dry seasons in 2010 and 2011 for various agri-cultural applications. The calibration and validation of the SSMmaps based on the fused images of AMSR-E and MODIS yieldedsatisfactoryagreementwith in situ groundtruthdatapatternwise.  Index Terms— Advanced microwave scanning radiometer onearth observing system (AMSR-E), leaf area index (LAI), mod-erate resolution imaging spectroradiometer (MODIS), normalizedmultiband drought index (NMDI), surface soil moisture (SSM). I. I NTRODUCTION S OIL MOISTURE is one of the most significant variablesin meteorology, climatology, hydrology, and ecology. Theimportance of soil moisture information has been extensivelydocumented by many hydrological and agricultural studies[1], [2]. Monitoring soil moisture can help predict potentialdrought impacts, which is useful for crop yield forecasting and Manuscript received October 26, 2013; revised February 28, 2014; acceptedJuly 30, 2014. Date of publication September 15, 2014; date of current versionJanuary 06, 2015.C.-F. Chen is with the Center for Space and Remote Sensing Research andthe Civil Engineering Department, National Central University, Jhongli 32001,Taiwan (e-mail: cfchen@csrsr.ncu.edu.tw).M. C. Valdez is with the Civil Engineering Department, National CentralUniversity, Jhongli 32001, Taiwan (e-mail: 973402601@cc.ncu.edu.tw).N.-B. Chang is with the Department of Civil, Environmental, andConstruction Engineering, University of Central Florida, Orlando, FL 32816USA (e-mail: nchang@ucf.edu).L.-Y. Chang is with the Center for Space and Remote Sensing Research,National Central University, Jhongli 32001, Taiwan (e-mail: lychang@csrsr.ncu.edu.tw).P.-Y. Yuan is with the Geospatial Informatics and Technology Laboratory,Taipei 235, Taiwan (e-mail: peiyao@hotmail.com).Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/JSTARS.2014.2347313 improvements of food management [3], [4]. In addition, there isa salient need for measuring spatiotemporal soil moisture distri-butions, particularly for climate predictions at the global scale[5]. Soil moisture is also of great importance in the hydrologicalcycle [6], [7] and all aspects of physical geography [8].The central region of Honduras, Guatemala, and Nicaraguahas an elevated mountain system. Such a landscape creates aclimatological contrast between the humid Caribbean regionsusceptible to floods and the long dry season of the Pacific coastof the Central America isthmus that provokes intense droughtin this populated region [9]. Drought is a frequent phenomenonand has a strong impact on basic food demand and availabilityof water for human consumption [10]. For these reasons, themonitoring and mapping of soil moisture gains societal rele-vance, particularly over the dry seasons for agricultural produc-tion assessment.Surface soil moisture (SSM) is known to be highly variablespatially and temporarily at all scales. SSM monitoring and pre-dictions can be carried out using several techniques, such as  insitu  measurements, remote sensing techniques, and hydrologicmodeling [8]. The use of   in situ  soil moisture measurements isconstrained by the demand from global or regional scale dueto spatial disadvantages [6], [11]. Using point measurements of soil moisture at the local scale is also inefficient and expensive[12],[13].Toovercomethisproblem,remotesensing-basedsoilmoisture estimation and mapping at the regional scale provide abetter understanding of the SSM dynamics with respect to spa-tiotemporal soil moisture patterns [14]. Remote sensing is alsoconsidered a good alternative because it averages small-scalevariability,providingthecapabilityforholisticanalysisandpat-tern recognition [6], [15]. Since 1980s, remote sensing tech-nologies have been used to produce more reliable soil moisturemaps with spatial and temporal variations [4], [11], [16]–[19].One of the most efficient approaches to characterizing SSMfrom space is microwave remote sensing [15], [17], [20]–[24].Microwave sensors are ideal because of their ability to detectEarth’s hydrological parameters such as soil moisture [25].The comparative advantages of microwave sensors over opticalsensors for characterizing soil moisture are related to the all-weather imaging capability, cloudy day performance, and watervapor and aerosol presence [25], [26]. The advanced microwavescanning radiometer on earth observing system (AMSR-E) isone of the most commonly known microwave remote sens-ing sensors. Since it has been successfully deployed on thesatellite platform of Aqua [21], the use of AMSR-E data for 1939-1404 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications standards/publications/rights/index.html for more information.  CHEN  et al. : MONITORING SPATIOTEMPORAL SSM VARIATIONS 4341 hydrological and geophysical mapping-modeling at global andregional scales has been widely acknowledged [27]–[30]. Forinstance, Zhang  et al.  [31] proved the usefulness of passivemicrowave remote sensing for observation of near SSM; how-ever, the use of AMSR-E for local scale soil moisture mappingis limited because of its coarse spatial resolution (i.e., 25 km)[32]. Furthermore, the use of active microwave remote sens-ing for geophysical or hydrological purposes has an increasinginterest. In this sense, the use of synthetic aperture radar (SAR)for soil moisture applications [33], [34], disaster monitoring[35], hydrological purposes [11], [36], and other applicationsis of great interest and has been extensively studied withpromising results.Optical remote sensing can also be used for assessing SSMwhen microwave remote sensing images with better spatial res-olutions are too costly to acquire. Many studies demonstratedthat SSM can be assessed with the use of optical remote sensing[1], [4], [16], [37]. Several relevant indexes for monitoring soilmoisture were derived with the aid of various optical remotesensing images across the wavelengths of visible, near-infrared(NIR), and shortwave infrared (SWIR) [18], [38]. Among them,the normalized multiband drought index (NMDI) is estimatedusing three wavelengths, including the one in the NIR centeredapproximately at  0 . 86 µ  m and the other two in the SWIR cen-tered at 1.64 and  2 . 13 µ  m [38]. Several studies applied NMDIto test its suitability for monitoring drought, soil moisture, veg-etation moisture content, and forest fire risk [3], [30], [39], [40].NMDI derived from the moderate resolution imaging spectrora-diometer (MODIS) data provide drought information and forestfire risk with finer resolution that can be up to 1 km. When com-pared with other indexes, NMDI exhibited comparative advan-tages for assessing soil moisture conditions in areas with lowvegetation density [38], [40], [41].Despite the advantages provided by optical remote sensingto monitor soil moisture with a better spatial resolution, opticalsensors cannot function well in cloudy days and cannot pene-trate vegetation canopy. The signals collected by optical sen-sors are highly diminished by the Earth’s atmosphere [16]. Inany circumstance, both microwave and optical remote sensingimageries have pros and cons that can affect the outcome formonitoring soil moisture regionally and locally. To develop acost-effective SSM mapping strategy, data fusion of imageriesbetween NMDI and AMSR-E was developed in this study toimprove temporal and spatial resolutions of the final remotesensing product [42]–[45].Even though fusing remotely sensed multisource data pro-vides comparative advantages, the task is challenging due tothe complexity of the spatial footprints, repeat cycles, and spec-tral difference within the input data [46]. Research with datafusion from multiple sources has been of great interest [47].Many studies have carried out data fusion to obtain an improvedproduct that can meet specific requirements. For instance, someapplied multisensor data fusion for hydrologic analyses [44],[48]. In specific, SSM products derived from microwave remotesensing were fused with optical images to obtain an improvedproduct with higher spatial resolution [25], [33], [49]; however,the methods vary depending on its objectives and the territorialcoverage of a study region.Theobjectiveofthisstudyistodemonstrateauniquecascadedata fusion technique in both temporal and spatial contexts formonitoring spatiotemporal SSM variations in Central AmericaIsthmus region at 1-km spatial resolution and 8-day tempo-ral resolution. By fusing high-resolution NMDI derived fromMODIS data and low resolution AMSR-E SSM product col-lected by Aqua across the dry seasons in 2010 and 2011, thequality of the SSM maps can be of great enhancement becauseit uses an index designed specifically to monitor soil droughtcondition. In specific, the method links SSM with the MODISleafareaindex(LAI)datasetaddressingthedisturbanceofSSMmeasurements due to high-density vegetation cover [38]. TheSSM maps with higher resolution generated in this study canlead to the improvement of agriculture planning and forest firerisk assessment providing an early warning system in regionshighly vulnerable to droughts and forest fires.II. S TUDY  A REA The Central American Isthmus is formed by Guatemala, ElSalvador, Honduras, Nicaragua, and Belize (Fig. 1), totaling anarea of   394 881 km 2 . The rainfall in Central America is alsovariable, depending on the wind direction and the position of the tropical and intertropical convergence zones. According tothe updated Koppen-Geiger climate classification for the 20thcentury [50], Central America has a predominantly tropicalequatorial humid climate, with tropical dry winters; the east-ern areas are rainiest due to higher altitudes. The central regionof Guatemala has temperate weather with dry winters and sum-mer rains. An elevated mountain system in the central part of Honduras, Guatemala, and Nicaragua [51] creates a climato-logical contrast between the humid Caribbean region, whichis susceptible to floods, and the populated Pacific coast of theCentral American isthmus, which has a long dry season thatprovokes intense droughts [9]. The economic activity in Cen-tral America is mainly based on agriculture, and many peoplerely on subsistence farming [52]. Recurrent drought affects theproduction of basic grains, which affects the demand for humanconsumption [10].III. M ATERIALS AND  M ETHODS This study aims to develop a cascade data fusion method tohandle a complex data fusion over different scales. The pro-posed cascade data fusion was confirmed by using the AMSR-ESSM microwave observations and the NMDI observationsderived from the multidate MODIS surface reflectance data andLAI MODIS data via the development of linear and nonlinearrelationships among these datasets. Such relationships can pro-vide integrated statistical information and bridge the gap of data fusion among NMDI, LAI, and AMSR-E soil moisturesatellite datasets acquired through two or more sources withdifferent scales [38]. Statistical analysis was used to mergethe datasets having an observed linear relationship toward animproved SSM product with a higher spatial resolution. Fur-thermore, a threshold was identified to eliminate the regionsin which there is no response of NMDI to SSM changes athigher levels of LAI. Interpolation method was used to account  4342 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 11, NOVEMBER 2014 Fig. 1. Study area including the five countries of Central America: Nicaragua, Honduras, El Salvador, Guatemala, and Belize. for unknown errors to the SSM estimation. In this paper, thecommon input variables for the interpolation approaches (nor-malized difference vegetation index, albedo, land surface tem-perature, etc.) are replaced with NMDI over different scalesbased on the cascade data fusion concept. After the interpo-lation procedure was applied, SSM with finer spatial resolu-tion can be estimated for the whole study region. Followingthis “cascade data fusion” philosophy, the best fitted statisticalmodel was found to estimate the SSM to account for delicatedrought impact assessment (Fig. 2).The use of statistical methods for data fusion provides manyadvantages [53], [54]. The proposed method is computation-ally efficient and does not require ground-truth data for SSMestimation at a regional/local level, thereby making the algo-rithm practical and operational. The working flowchart in Fig. 2follows five main steps to build up this operational systemincluding: 1) data collection; 2) data preprocessing; 3) LAIthreshold estimation; 4) model calibration and validation; and5) production of high-resolution SSM estimation at 1-km spa-tial resolution over the dry seasons in 2010 and 2011 fordemonstration.  A. Data Collection Three types of remote sensing data were used in this studyincluding: 1) MODIS optical images with 8 day and daily tem-poral resolution (e.g., NMDI); 2) MODIS LAI images; and3) AMSR-E SSM microwave data. Images which have lowcloud coverage were acquired during dry seasons from Januaryto April of 2010 and 2011, addressing the most critical droughtconditions in the Pacific region. They are introduced below. 1) MODIS/Terra Surface Reflectance Data:  TheMODIS/Terra surface reflectance 8-day L3 Global 500 m(MOD09A1) data collected during dry seasons (January–April)of 2010 and 2011 were acquired from the National Aeronauticsand Space Administration (NASA), which were used to calcu-late NMDI. Calculations of NMDI are carried out using NIRband 2 ( 0 . 841 − 0 . 876 µ  m), SWIR band 6 ( 1 . 62 − 1 . 65 µ  m), andSWIR band 7 ( 2 . 10 − 2 . 15 µ  m) simultaneously. The productcontains the best possible pixel observation during an 8-dayperiod selected on the basis of high observation coverage,low viewing angle, low cloud coverage or absence of cloudshadow, and aerosol loading [55]. The MODIS Terra instru-ment was used because the Aqua MODIS instrument band 6( 1 . 62 − 1 . 65 µ  m) has experienced performance problems sincelaunch in 2002; 15 of the 20 detectors were either nonfunctionalor noisy during our study period [56]. In addition, the MODISTerra/Aqua Surface Reflectance Daily L2G Global productwith 500-m and 1-km resolution (MOD09GA) was used to testthe capability of the method for downscaling the SSM to 1-kmspatial resolution with daily temporal resolution for emergencyresponse purposes. MOD09GA provides MODIS band 1–7daily surface reflectance at 500-m resolution [55] and the bands2, 6, and 7 were used for calculating the NMDI. 2) MODIS LAI Data:  LAI is defined as one side green leaf area per unit ground area in broadleaf canopies and as the pro- jected needle leaf area in coniferous canopies [57]. LAI is acrucial parameter in remote sensing as it can provide a basisfor identifying phenology of vegetation from satellite images[58]. LAI is also an important variable for hydrology, ecology,and climate modeling [59]. In this study, MODIS FPAR/LAIdata (fraction of absorbed photosynthetically active radiationthat a plant canopy absorbs for photosynthesis and growth inthe 0.4–0.7 nm), which is LAI 8-day Global SIN Grid V005at 1-km spatial resolution (MOD15A2), were used to separatehigh- and low-density vegetation areas. The FPAR LAI MODIS  CHEN  et al. : MONITORING SPATIOTEMPORAL SSM VARIATIONS 4343 Fig. 2. Flowchart of this study for cascade data fusion of microwave and optical remote sensing images to generate multitemporal SSM distributions with 1-kmspatial resolution. land product files were used in this study, each of which con-tains four scientific datasets (SDSs) with spatial resolution of 1 km [60]. 3) AMSR-E SSM Data:  AMSR-E is a microwave radiome-ter for six frequency bands from 6.9 to 89 GHz which scans theEarth’s surface and acquires radiance data of the Earth’s sur-face [61]. The AMSR-E instrument provides a wide variety of products for research, including sea surface temperature, sea iceconcentration, snow water equivalent, SSM, surface wetness,atmospheric cloud water, and water vapor [61]. These productsalso have different levels of processing including L1A, L1B,L2, and L3; the L3 data are produced by averaging temporar-ily and spatially the L1B (brightness temperature) and L2 geo-physical parameters data [61]. In this study, the AMSR-E/AquaDaily L3 (Level 3) SSM ( mg / cm − 3 ), Interpretive Parameters,and QC EASE-Grids V002 25 km AE Land3 datasets wereused as the reference [21], [62], [63].  B. Data Preprocessing The preprocessed MODIS data include the LAI/FPAR prod-uct and the MODIS surface reflectance data. Since MODISimages use the Sinusoidal projection (Equal Area Projection),both datasets were converted to  universal transverse mercator  (UTM) coordinate system using the MODIS Reprojection Tool(MRT). The MODIS surface reflectance images were resam-pled during the coordinate transformation process using thebilinear interpolation resampling method. Resampling was car-ried out only for the MODIS surface reflectance images tomatch the spatial extent of the MODIS LAI product, which  4344 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 7, NO. 11, NOVEMBER 2014 Fig. 3. Samples of NMDI response to AMSR-E soil moisture change at different LAI intervals used for each regression: (a) January 25, 2010; (b) February 2,2010 in the Central America region. has a spatial resolution of 1 km. Finally, with the aid of vec-tor data from the Central American region, water bodies weremasked out, and pixels covered by clouds were eliminated byusing an automatic cloud detection method based on spatial tex-ture analysis [64]. C. LAI Threshold  NMDI is an index that performs well at monitoring dry soilstatus when vegetation density is low [30]. The NMDI was usedas the high spatial resolution reference of the soil moisture con-ditions to be included in the statistical model for SSM estima-tion in this study. In particular, it is possible to test NMDI’sperformance using SSM data collected in areas with differentLAI values [39]. Since it was observed that the NMDI stopsresponding to soil moisture increase when LAI is high, it istherefore necessary to estimate a threshold for LAI in orderto reflect such a disturbance [39]. To develop this threshold,NMDI was estimated using the MODIS surface reflectancedata at first. Then, the response of NMDI to SSM change wasanalyzed using a linear regression analysis based on AMSR-ESSM data and NMDI data in order to retrieve the behavior of NMDI in response to SSM changes at different LAI intervalsdescribed by satellite datasets. Results were plotted and ana-lyzed to identify proper threshold for eliminating the negativeimpact of dense vegetation areas in the final step of the datafusion.The data preprocessing for this dataset has two stages,including the coordinate projection transformation and the tem-poral resolution matching. Raw data of the global SSM prod-ucts of AMSR-E have an Equal-Area Scalable Earth Grid(EASE-Grid) projection and coordinate system. As this is aglobal dataset, the images of interest need to be subset to matchthe Central American study area at first, and the images werethen converted to  UTM   projection coordinate system for finaldata fusion. AMSR-E provides data generally on a 2-day basisbecause of the ascending–descending node acquisition patternthat creates a 2-day gap in the study region [61]. In order tomatch the temporal resolution of MODIS datasets, the imagesare averaged for an 8-day period. The temporal resolution of the final data fusion product presented in this study is 8 days;however, for emergency response, temporal resolution had bet-ter reach up to 2–5 days by using a different set of MODISsurface reflectance images with daily temporal resolution. Forthis purpose, AMSR-E data are averaged accordingly. In thisstudy, a product with 8-day temporal resolution is preferred toillustrate the downscaling method in a less cloudy condition.This scheme contributes to the so-called “cascade data fusion”in the temporal context. 1) NMDI:  NMDI is an index that can be used to extract soilmoisture information because a linear relationship is observedbetween both variables. This means an increase or decreaseof soil moisture is linked with a reduction of NMDI propor-tionally; hence, this variation is useful to extract soil moistureinformation [18], [38]. The index, however, is only suitableto monitor soil moisture in low-density vegetation areas. Forthis reason, it is necessary to analyze the behavior of NMDI inresponse to the AMSR-E SSM change at different LAI intervalsusing satellite data. NMDI is an index proposed by [38] and isdefined as NMDI   =  R 0 . 86 µ m  −  ( R 1 . 64 µ m  −  R 2 . 13 µ m ) R 0 . 86 µ m  + ( R 1 . 64 µ m  −  R 2 . 13 µ m )  (1)where  R 0 . 86 µm ,  R 1 . 64 µm , and  R 2 . 13 µm  are the apparentreflectances observed by a satellite sensor [38]. NMDI uses the 0 . 86 µ  m channel as the reference and the difference betweentwo liquid water absorption channels centered at 1.64 and 2 . 13 µ  m as the soil and vegetation moisture sensitive band [40]. 2) NMDI Response to AMSR-E Soil Moisture Change at  Different LAI Intervals:  Linear regression analysis was usedto further confirm if there is a linear relationship betweenNMDI and AMSR-E SSM data. The purpose of this practiceis to identify the sensitivity of NMDI in response to AMSR-ESSM changes at different LAI values. If the slopes associatedwith those resulting regression lines may show a linear patternbetween NMDI and AMSR-E SSM, when LAI values are low,the fidelity of this proposed cascade data fusion in Fig. 3 maybe confirmed [38]. In other words, the microwave satellite data
Recommended
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x