Overview:

Current techniques allow interpretation of hyperspectral remote sensing data to determine standard land cover types (vegetation, water, soil, and rock type) without significant field verification. However, in order to determine specific conditions about land cover types such as vegetative health, subsurface contamination or specific water quality criteria, ground-truth data collection is still a necessity. Once it has been determined that "ground-truth" or validation data are required, the following discussion and methodology can help to design the sampling plan.

In general, the collection of five types of "ground-truth" or validation data are recommended:

Atmospheric Conditions
Dark and Light Calibration Targets
Surface Water (Where Present)
Vegetation
Soil, Bare Ground, Rock Outcrop

Atmospheric conditions: ­ These data are used to correct the hyperspectral image for conditions in the atmosphere that intercept incoming solar radiation, thereby affecting the intensity or frequency of reflected energy signals. It is preferable to collect this data on the date of the collection of the image, however, as an alternate, atmospheric data collected close to the date of collection under similar atmospheric conditions and at approximately the same solar time, can be substituted. The standard time to collect hyperspectral images is at or near solar noon, with the window being from about 2 hours prior to 2 hours after noon. The hyperspectral image will not be obtained during cloudy or inclement weather. The surface water and vegetation spectral data also should be collected at solar noon plus or minus about 2 hours. This time may be extended to 3 hours plus or minus for days close to the summer solstice. Depending upon latitude of the study site, collection early in the spring and late in the summer can become infeasible due to low sun angles.

The atmospheric conditions of interest are:

Temperature
Humidity
Haze or aerosols
Wind Direction and Speed
Incident Solar Radiation

A recent ground-truthing effort included erecting portable weather stations at three refineries. The station operated continuously (24hrs) during the 3 to 5 day period over which the vegetation and surface water sampling was conducted. The weather data record identifies any changes in conditions during the field measurement period. In addition, solar radiance can be measured directly during the overflight, using a cosine diffuser attachment on the spectrometer and the aperture oriented vertically up, and no obstructions (e.g., buildings or trees) in the field of view above 35 above the horizon.

Dark and Light Calibration Targets: Acquiring spectral measurements of uniform bright and dark calibration target areas within a site is recommended. The minimum size of these target areas should exceed the pixel resolution of the imaging instrument, preferably sizes equal to 9 to 25 pixels or more in extent. These target areas should be on flat ground, relatively homogeneous (little splotching or color variance within the area), and as close to white or black, respectively, as possible for a given site. Examples of the types of possible targets that can be used are asphalt, cement, sand, a flat aluminum roof, or bare dirt areas. It is best to acquire these as close to the flight time (or conditions at the time of the flight) as possible. The reflectance of these dark and light targets, and their known spectral response, are used to calibrate the hyperspectral imagery and to correct for atmospheric influences. Typically, dark or new asphalt makes a good dark calibration target and cement pads (e.g., a helicopter landing pad or a large roof area) are used for the light calibration target.

Surface Water: ­ Where surface water bodies are of sufficient size and interest, ground data should be collected to represent each type of water condition (representing varying water depths, temperature, relationship to potential source areas, fresh or salt water bodies, turbidity, etc.). Clear water absorbs most of the solar radiation in the bandwidths imaged by hyperspectral remote sensors. Reflectivity increases with the presence of solids, either inorganic or organic and the presence of certain dissolved species can also alter the hyperspectral image. Understanding the cause in different reflectivities and absorptions in surface water is important to the final interpretation of the image.

The surface water data to be collected includes the following, although some data will not be collected depending on the availability of nearby analytical services:

Georeference Point (GPS)
Incident Solar Radiation
Water Turbidity, Secchi Depth
Spectroradiometer Readings (Spectral Signature of Calibration Target and Water Surface at ~2 m)
Chlorophyll Content
Spectral Signatures Within Water Column at Various Depths
Spectral Signature of Sample Water Compared to Distilled Water Reference (Bench Test)
VOC, SVOC, Metal, Other Dissolved Chemical Analyses (Based on Project DQOs)
General Notes About the Location and Light Conditions

Vegetation: ­ Presuming that the analysis of interest is that of vegetative health/stress, species composition or plant community mapping, spectral measurements of the various vegetation types in the field will be required.

The spectral signature within a pixel of the image consists of an average of the reflectances of all materials within that pixel. For example, for a spatial resolution of 5 x 5 meter pixels, the spectral response for a stand of vegetation will consist of a combination of the spectra of all vegetation types and the soil, ground litter, etc., within the picture element. Ground-truthing consists of obtaining spectrometer readings (or samples) for as many of the categories/classes of vegetation types (monoculture and mixed species) as possible within time or access constraints. These readings are obtained above the canopy and should try to be representative of reflectances detected by the airborne or satellite remote sensor. For instance, these categories could include a grassy field, a stand of pine with only forest litter at ground surface, a mix of eucalyptus, live oak, vines, and chinese tallow, a salt marsh with cattails, a fresh marsh with mixed vegetation, etc. Each category should be sampled from above the canopy. Presently, it is also valuable to collect actual leaf samples for bench analysis of reflectance to provide end-points in the analysis (to answer questions e.g., "If the canopy were all leaves, it would look like the leaf spectra. To the extent it is composed of several materials, it will look like mixtures of them). The leaf, soil, and litter measruements are done in an office or lab and involves the use of a power stabilized known light source and calibration standard. It is similar to the bench water sample analysis mentioned above.

A further consideration is that of sample stratification; that is, whether to collect many readings in a single location (narrow-deep sampling) or to collect few samples at many locations (broad-shallow sampling). Site conditions or project goals may drive this decision.

Finally, if the understanding of surface or subsurface contaminant conditions is desired, spectral data should be collected in areas of known contamination as well as locations considered to be background or uncontaminated. Transecting across a known plume from contaminated to background might be a desirable approach.

Data types to be collected at each location (including bright and dark targets) includes:

GPS
Spectroradiometer of Calibration Target and Vegetation
Vegetation Sample (Leaf or Leaf Cluster)
General Field Notes Regarding Site and Light Conditions

Soil Types: ­ Soil measurements, as in vegetation spectral sampling, data collection should strive to represent the various types of soils or contamination conditions, which occur at a remotely sensed site. Soils are darker if they are wet, so at sites where water content varies, it is a good idea to measure both wet and dry soil conditions. Wet and dry soils, rock outcrop, and other geologic conditions may also warrant collecting location specific data collection, e.g., GPS. The data types will depend on project DQOs but may be the same as for vegetation sampling.

Field Equipment and Procedures

In general, the service provider or specialty consultant will provide the spectral measuring devices. These will be named here but the listing is not intended to be all inclusive nor will this document specify when these are not needed. Project DQOs and consultant expertise will dictate the necessity of any given instrument or support equipment.

Atmospheric Conditions

Equipment

Continuous monitoring weather station -- wind speed/direction, humidity, temp, total radiance, battery, data logger with computer link for download.
Spectrometer and cosine diffuser

Procedure

The weather station needs to be placed in an area which is completely open to the horizon in all directions (as much as practicable) and which is not influenced by very local microclimates (such as wind around a small hill or structure) or shadows from natural or man-made features. As this station may be left operating for several days, a secure location is preferred. If the hyperspectral image will not be obtained while the weather station is operating, it is desirable to obtain as much weather data for the imaging time as can be obtained from local sources (airport, National Weather Service, etc.). In many cases, the consultant performing some other aspect of the ground-truthing will also be gathering the atmospheric data. Set-up time for this station is generally less than 1 hour and the equipment will take two people to deploy.

A spectrometer fitted with a cosine diffuser is placed (pointed at the sky) in an area which is completely open to the horizon in all directions, away from shadows and sources of specular reflection. Readings are taken continuously immediately before, during and after an overflight.

Dark and Light Calibration Targets

Equipment

Spectroradiometer, Computer, Calibration Target, etc.
GPS
GIS-Based Site Maps, Property Maps, Surveyors Benchmark Maps, etc.
Aerial Photos
Data Sheets
Camera

Procedure

The bright and dark target areas will typically be sampled along one or more transects to help quantify variations in the target's reflectance. Suitable dark target sites include: smooth asphalt parking lot; clear water body, dark soil. Suitable light targets include: cement lot, flat, smooth aluminum roof; white sand. GPS readings will be collected at each location at which a spectrum is obtained. All suitable sampling locations should be identified by a reconnaissance tour prior to actual data collection.

The field data collection effort will be undertaken primarily between 10 a.m. and 2 p.m. local standard time, or within about 2 hours before and after local solar noon. Each sample site should be described in the field notes (sample data sheet attached) to enhance the subsequent interpretation.

Surface Water

Equipment

Motorboat, Large Enough for Two Technicians and One Operator
Cooler
Sample Bottles (1-Liter, Unpreserved, 3 Per Data Collection Site)
Distilled Water
Spectroradiometer(s)
Turbidity Meter
Secchi Disk, Tape Measure
GPS
Camera
Freezer (if Samples To Be Shipped for Analysis)
Data Sheets
Maps, Aerial Photos, etc.

Procedure

The number and locations of data gathering sites will be dependent on project DQOs and whether surface water of sufficient size is on or nearby the project location. General site selection criteria were specified in an earlier section.

The process at each sample location will take about 1 hour, with travel time between sites varying, depending on property site, access, etc. As an example, it took about one-half hour to get from sample site to sample site in San Francisco and San Pablo Bays around the Richmond Refinery.

It is preferable to have two sampling personnel (the expert consultant with a helper) and a boat operator who may have to operate the motor to maintain a nearly constant position in the water. The boat needs to be positioned so that one side constantly faces towards the sun (no shadows from the boat/people). The GPS location is obtained and three incident (spectroradiometer) light readings are collected as an initial step. These light readings consist of an open sky reading, one with the direct sun blocked (technician holding a shade object) and a final reading of just the reflected light from the technician and no shade. These three readings will be repeated to conclude the data collection at each site. The consultant makes general field notes including time of day, approximate wind direction, surface water roughness, haziness, etc.

Water clarity readings are obtained with the secchi disk (record depth below water surface that the disk is no longer visible) and a turbidity meter or equivalent. The secchi depth is used to determine the depths from which to record spectroradiometer data. The spectroradiometer is lowered into the water column with the light sensor facing skyward for readings at "X" multiples of the secchi depth, and again with the sensor facing down. These readings capture the water's transmission and backscattering of light, respectively.

Water samples will be collected by reaching into the near surface of the water column and filling a 1-liter plastic jar. One sample will be used for spectral response comparison with distilled water as the standard (bench/lab work), one for chlorophyll analysis (if specified by the DQOs), and one will be frozen for shipment to an analytical lab (also if specified).

Four photos will be taken ­ one directly upwards, one directly down (at the water surface), one at horizon in the direction of the sun, one opposite horizon.

Prior to repeating the incident solar readings, a second spectroradiometer will be used to obtain single, downward facing spectra, directly over the water. Prior to this water surface reading, a calibration (white) reading is obtained using a special, reflective target (one such material that meets National Bureau of Standards specifications is known as "Spectralon").

Vegetation And Soil

Equipment

Bucket truck, Man-lift or Other Method to Take Readings Above Plant/Tree Canopy
Spectroradiometer, Computer, Calibration Target, etc.
Cooler
GPS
Integrating Sphere ­ Bench Spectral Readings
GIS-Based Site Maps, Property Maps, Surveyors Benchmark Maps, etc.
Aerial Photos
Data Sheets
Camera

Procedure

The general procedures for obtaining ground-truth spectral readings for vegetation cover and open soil or rock are about the same. Slopes should be avoided, especially steep slopes, and locations that are smaller than the image resolution may not provide useful data. Prior to field activities, the project team needs to agree on whether the appropriate data are broad-shallow or narrow-deep, as defined earlier. These two nearly opposing sampling strategies will further narrow the choice of acceptable sample locations. Sampling sites with difficult access will potentially constrain the total number of sites sampled as data collection at a single site can take up to 1 hour if the strategy is narrow-deep. Broad-shallow sampling may only involve 10 to 15 minutes if set-up is minimal (i.e., no bucket truck or ground preparation needed).

GPS readings will be collected at each location at which a spectrum is obtained. All suitable sampling locations should be identified by a reconnaissance tour prior to actual data collection. It also helps if aerial photos are available for the project property under investigation. Equally valuable, if available, are site contaminant distribution maps if understanding contaminant impacts is a project objective.

The field data collection effort will be undertaken primarily between 10 a.m. and 2 p.m. local standard time, or within about 2 hours before and after local solar noon. Project teams should prepare equipment and mobilize into the field to be ready to sample at the beginning of this time window. After the day's field period, the team(s) will typically demobilize and conduct the bench spectral analyses.

Each sample site should be described in the field notes (sample data sheet attached) to enhance the subsequent interpretation. Examples of important site descriptions include:

Vegetation Type and Species
Monoculture or Mixed Community
Single Layer or Multiple Layer Community (Grass or Tree Over Grass)
Type and Distribution of Ground Cover (Open Soil, Litter, etc.)
Apparent Vegetative Health and Growth Stage
Disturbances
Pattern of Type Distribution ­ Sharp Contact Between Species or Age Classes, Other Patterns
General Topography
Time of Day and General Condition of Sky (Clouds, Wind, Haze, etc.)
Soil Conditions (Wet or Dry)

Field crews will attempt to dress in low reflective clothing in darker colors. If special PPE is required, such as flame retardant overalls, tan or dark blue are preferred over yellow or orange. Reflection from field equipment will also need to be mitigated, so that bright white bucket trucks or other vehicles should be parked at a distance or covered with dark cloth (especially important for the bucket itself). The spectroradiometer will generally be attached to the boom truck, with an operator and field technician on board. If a grass field or meadow is being sampled, the technician will likely walk a transect across the field.

Spectral data collection will consist of a calibration using the Spectralon target, and multiple spectral readings above each sample site. The calibration will be repeated several times during the sample period to establish changing light conditions or instrument drift. The field technician will have calculated the appropriate height to collect data above the target sites based on instrument parameters. Readings will be taken so as to avoid shadows, so that field equipment and vehicles will be positioned accordingly.

Each sampling site will be georeferenced (GPS), generally by the second field technician, who will also collect and store appropriate vegetative samples. Larger sites (meadow, large tree, building) and roads may be georeferenced at their circumference in order to further "fit" the image to actual features on the ground.

Additional Considerations

Safety Training ­ Many industrial sites have site-specific safety training requirements. Sampling programs may need to allow additional time to meet these requirements.

PPE ­ Industrial sites often require specific clothing, headgear, etc., which should also be accounted for in the program plan. Sampling crews should use sunscreen, hats, or other protection to avoid sunburn, windburn, etc. Depending on location and time of year, ground crews should be especially aware of biologic hazards such as fire ants, snakes, etc.

Radios ­ Depending on property size and number of field teams, some form of direct communication between teams and between the facility operators and these teams may be desirable.

Site Guide- It is often preferable to have a person familiar with the site act as a guide so that field teams can focus on data gathering and not on getting around the property.

Helicopter ­ In some cases, direct access to the ground may not be possible, therefore, a helicopter may allow spectral readings to be obtained above the vegetative canopy. This will require special arrangements that should be developed on a case-by-case basis.

Drinking Water, Food, Supplies ­ All teams should bring sufficient drinking water to the field to sustain a 6-hour effort. Meals may need to be ordered in advance so that samplers can maximize data collection during the solar-noon window. It may also be desirable to have a support person available to deliver food, supplies, etc., to the sampling crews.

Large Equipment ­ Pickup truck, boom trucks, boats, etc., may not be available at all properties; therefore sampling teams should be prepared to provide these through rental organizations. Operators familiar with the equipment and possibly safety trained to the same level of awareness as the local plant operators will also be necessary.

Standard Field Spectral Data Collection Protocol

Adhering vigorously to an agreed upon protocol, and collecting measurements in a consistent and thorough manner will minimize variability associated with the actual process of data collection (and recording). It is important to take into account the possible influence of systematic bias being introduced, either inadvertently incorporated into the sampling protocol, or as a result of unconscious bias on the part of the fieldworker and data collector. This influence can be aggravated in the case of multiple persons collecting datasets overtime. Although not all situations can be anticipated, accounting for possible bias with a set of predetermined criteria, resulting in a set of rules for making judgement calls (clearly iterated within a written protocol), will substantially improve data quality.
A specific data collection methodology must be tailored to project and site specific criteria, i.e. research questions, ecosystem type, vegetation types, site accessibility, anticipated analyses, and desired end-products and the time and resources available. In general, however, field data collection for hyperspectral remote sensing consists of several categories of activities, each with it own set of criteria for ensuring data quality:

Pre-Fieldwork Planning and Preparation:

1.) Site selection
In most situations, selection of both the site and the main areas of research interest will have been already chosen by the time planning for fieldwork begins. However, within most field environments there is both a wide latitude in specific decisions and significant limitations that determine the selection of the specific criteria for the project. A full scene is too large and complex to fully sample. Therefore, tradeoffs must be made between number of replicates and number of classes that can be measured within the time frame allowed. The goal of the field work is to identify sufficient information on the ground that will support the image interpretations - not to duplicate them. Not all locations within a site are easily accessible. As a general rule, unless there is something critical to the success of the project located in a difficult to access site, it is better to opt for more easy-to-get samples that represent the major cover types of interest for the project. If the study involves point sources of contamination or gradients in a site condition, orienting data collection to measure transects (points along a line) that pass across the "problem area" are very useful. This gives you references for the problem and the "control" condition and some intermediate states. In other cases, finding "pairs" of samples with the "control" and "experimental" conditions may be feasible. This type of sampling works well if you have some known areas that express the condition you want to evaluate in the image data.
It is useful, when possible, to make a reconnaisance of the area before the start of the actual fieldwork, taking note of potential sampling sites. A "wish" list of sampling sites can be compiled and marked on the map. This list can then prioritized, for example with sites rated as "essential" or "important" or "interesting". It is sometimes usefule to outline a set of goals and priorities for the particular fieldwork exercise. Issues that need to be addressed at this time include:

1.) Overall purpose of the fieldwork:
A.) Vegetation/ landuse mapping
1.) Major landuse types / vegetation communities
B.) Plant stress detection
1.) Control sites
C.) Detection or mapping of specific interest, e.g. contaminants
1. ) Control sites
D.) Change detection
1.) Seasonal effects
E.) Imagery analysis and groundtruthing
1.) Timeliness of data collection
F.) Georeferencing of satellite or airborne imagery

2.) Specific data to be acquired:
A.) Hyperspectral data collection
1.) Canopy reflectance
2.) Leaf samples
3.) Soil or bare ground
4.) Litter or non-photosynthetic biomass
5.) Calibration sites - e.g. asphalt or cement
6.) Ambient calibration measurements, e.g. solar radiance
B.) Plant or litter samples
1.) Which analyses will be performed
a.) Hyperspectral
b.) Biomass weight (wet)
c.) Lab analysis (chemical)
2.) How long before samples arrive at lab
a.) Need ice and ice chest
b.) Sample bags
3.) Sampling protocol
a.) How much material is to be collected
b.) Which plants / leaves
c.) How chosen
d.) Labeling
C.) Soil samples
1.) Which analyses will be performed
d.) Hyperspectral
e.) Biomass weight (wet)
f.) Lab analysis (chemical)
2.) How long before samples arrive at lab
c.) Need ice and ice chest
d.) Sample bags
3.) Sampling protocol
a.) How much soil will be collected
b.) Surface/ subsurface / profile
c.) Spot or pooled samples
D.) Photographic samples
1.) Site Description
2.) Plant ID
3.) Change Detection

3.) Major vegetation communities / species present within the study area:
a.) Which need to be sampled
b.) How will they be sampled
c.) Sampling intensity

4.) What are the major environmental gradients within the site:
A.) Elevation/topographic gradients
B.) Moisture gradients
C.) Contaminant or pollution effect gradients
D.) Landuse gradients

5.) Are the sites of interest accessible?
A.) Can they be sampled from the ground?
B.) Are sites near to road representative?
C.) Will a ladder/bucket truck/crane/helicopter be required?

Based upon this basic outline and the annotated map, a sampling plan should be delineated. It is important to plan for adjustments when time runs short. Further, if weather conditions turn out to be a factor, several scenarios can be shifted to allow for sampling of priority sites during clear conditions . If it's a clear sunny day, and there is a possiblity the next might not be, it's a good idea to get the priority sites done first. That way, in case it does rain for the next two weeks, you will still have gotten the bare minimum. Usually, essential data for imagery analysis will include indentification, delineation and sampling of both dark and light calibration sites.

Pre-fieldwork Preparation

Pre-fieldwork prep and organization are a particularly important component of a successful field campaign. Once in the field it can be difficult or impossible to cope with or fix problems, or improvise solutions for forgotten items. A lack of organization in planning for an extended campaign, especially if travel is involved, may easily compromise the effectiveness and efficiency of data collection. New equipment should be assembled, tried out, used, and results analyzed before being taken out into the field. Likewise, all new personnel should be trained in the protocol, and the use of the instruments before going out to the field. Among essential items which need to be organized and which can substantialy improve a field campaign are:

1.) Maps and Site Description
A.) Map of area and proposed sampling sites
B.) Have a copy that can be marked and written on in the field
C.) Check georeferenicng, projection, datum
2.) Plant Indentification
A.) Make list of dominant/abundant vegetation types and species with both common and scientic names
B.) Assign each species/community/landuse a code e.g. "pickerelweed" could be assigned "pweed" or, based on its scientific name, Pontederia cordata would be "pon cor"
C.) If detailed species id is required, than plant id books and taxonomic keys might be useful if taken along. Plant samples can be taken or pressed for later id at a herbarium.
3.) Equipment list
A.) Comprehensive list of all equipment to be taken out into the field.
B.) Each piece of equipment should be listed and all accessories/ cables/ batteries/ etc. itemized.
C.) A list of all pieces of equipment which go into a carrying case should be included (taped) within the case
D.) If more than one person is in charge of equipment, then responibility for specific instruments should be clear delineated
4.) Charge batteries / spares
A.) Are batteries charged
B.) Do batteries need to be charged in the field / car
C.) Is there a place to charge batteries overnight

5.) File Naming Conventions and Data Dictionaries
A.) Adopt a file naming convention and labeling scheme for all data to be collect before going out into the field.
B.) All files names should be unique, so there can never be any confusion
C.) File names should be systematic and explanatory
D.) File names can coded, e.g., if multiple sites are to be sampled, each over several days, the first character in the name would refer to the site, the second refers to the day of sampling, and third refers to the number of the site sampled that day. For example, TA1.000 would refer to a the first sampled reading taken at the Texas site, on the first day. Likewise, MC7.000 would refer to the seventh site on the third day of sampling at the Mississippi site.
E.) GPS's and some data loggers allow for data dictionaries to be set up to easily record predetermined features

6.) Synchronize times
A. ) Set and synchronize the time on all instruments which record time
1.) Computers
2.) Cameras
3.) Spectrometers
4.) GPS (this is done automatically, and should be used as the reference time)

Data Sheet

It is easy to forget to record essential data when you are in the field. This is especially true when several people are working and assuming someone else is recording some essential piece of information. Pre-determined data sheets are critical to obtaining a complete record. It is useful to make sure data records are cross-referenced (e.g., recording the GPS file name with the spectral data that it goes with) even if this information is being entered electronically (for example in the GPS data dictionary). Recording time and date stamps is very helpful for cross referencing files and data. Sometimes recording errors can be fixed later if this information is co-recorded. It is sometimes useful to bind data sheets together into a data book. This allows for a data from one site to be in one place, and keeps individual sheets from getting misplaced. Depending upon specific requirements data sheets should include:

1. Site Description
When a specific location is visited in the field, it is important to have a general description of the characteristics of the area. Things like, whether it is a monospecific or mixed community, whether it is a one-layer crown (e.g., shrubs or grasses) or a multiple layer community (e.g., a tree layer and an understory grass layer). Are all the vegetation types in the area in the same stage of phenology? (i.e., any dormant?). What is the general type of vegetation, it is homogeneous or heterogeneous (and in what way), what is a typical proportion of ground cover (or range). Is it a single soil type or heterogeneous soils? What is the condition of litter (surface, standing) and how much? Is there some anomoly in the vegetation distribution in the area? E.g., low or no cover, dead or scenescent vegetation, evidence of disturbance, recent or old? Sharp discontinuity in the type of vegetation or age class or some other distinct pattern? It also helps to know a description of the local topography. Writing down complete observations is useful. Especially noting unusual conditions helps to interpret field information later. Things like weather conditions (partial clouds? Wind?), pure stands or mixed species?, more or less standing litter in the crown? Flowering? Wet soil? Canopy pruned? Insect damage? It often helps to have photos of the objects that were measured and landscape shots of the general area. These need to be recorded where they were taken, what was the object and for landscape views, what orientation is shown. Physical parameters such as slope or aspect should be noted or measured.

2.) Spectrometer and Spectrometer Settings:
Record which instrument is recording the data, e.g. ASD-FR, GER 2600 or GER 1500. Also record instrument settings and whether using fiber optic, diopter, or cosine diffuser. This information is recorded in the GER 2600 files.

3.) File Name
This is the actual file or reading taken by a specific instrument. For example, for the GER 2600 spectrometer, a new file is given a name based on the site (according to the code explained above, e.g. TA1, meaning the first site on the first day in Texas). The GER then assigns file number sequentially to that file name, starting with xxx.000 (i.e., TA1.000). If this file is being recorded onto a computer, like the GER 2600 does, then is important to locate all files within one directory, and to organize a directory structure

4.) Reference or Target File:
Record whether this is a reference (calibration) reading taken off the white reference standard, or a target reading. This is done automatically if using the GER 2600.

5.) Corresponding GPS File Name
Most likely, this will be the corresponding GPS file where the reading recorded in file name is being recorded as a feature.

6.) Date and Time of Data Collection
Recording of time can very helpful for cross-referencing files and data. If many reading are being taken rapidly, this could be recorded as start time, or recorded every five reading, for example.

7.) Vegetation / Landuse / Calibration Type
Many times sampled vegetation will be a mixture of several to many species and/or bare soil or litter. It can be useful to characterize vegetation abundance (within the field of the instrument) as primary, secondary, tertiary or more. Generally, canopy cover or abundance is expressed as a percentage of total area cover by a particular species or vegetation type. To expedite data recording, species codes (as explained above) can be used. Species found in the field by not previously assigned a code can be identified and assigned a code in the field, and this information duly noted on the species list.

8.) Physical samples taken concurrently at this sample site:
Plant or soil samples should be labeled with a similar, but uniquely identifying type code. All sample bags and containers should be labeled, and this label recorded on the data sheet (along with corresponding GPS file, if there is a separate one).

9.) Photos recorded at this site:
It can be helpful in identifying corresponding photos if these are recorded at the time they are taken, generally by roll number and photo number. Digital cameras will have a file name for each picture.

10.) Conditions:
Record prevailing weather and atmospheric conditions, and specific conditions of the samples, e.g. wet, disturbed, senescent, flowering, etc.

11.) Comments:
Record any special conditions, or important information, or unusal performance of the instrument. Examples include: downstream from spill, abundant wildlife, presence of endangered species, computer is acting up intermittently, or noting problems, "Spectralon reference panel fell in the mud but we cleaned it up."

Before leaving the site, examine data sheet for missing information.

Conditions suitable for spectral measurements

There are many weather and site factors that affect spectral measurements, so it is best to minimize as much as possible their contribution to spectral measurements.

1.) Sun Angles:

The biggest factor that affects the reflectance is the sun angle. It will have a large impact on the overall albedo (brightness) of the spectra and will dominate the "minor" types of spectral change that you will be trying to measure. Sun angle varies with the time of day and day of year. This is the reason for acquiring field spectra (and image data) near solar noon. The sun angle changes slowly around noon, making it possible to collect comparable field data over several hours, centered around noon. The general rule is to try to limit data collection to +/- 2 hr of solar noon. But often, the cost of field crews makes this limitation impractical. If you can make an assumption of random scattering of light off of the surface, then a cosine correction can be applied. Unfortunately, plant canopies often violate this assumption. So, you need to consider whether this will cause trouble in interpretations or not. Sometimes, it is possible to re-measure a canopy at two times to determine the magnitude of change, thus putting an error estimate around the measurements. Sometimes it is worth making some measurements at the time of the sensor crossing (e.g., 9:30 am if you are using TM) to determine how large a difference it makes between your field "ground truth" and the image data.

This problem is what makes seasonal data hard to rigorously compare. It is easy to tell some major land use change (e.g, someone cut down all the trees). It is hard to determine "stress" between time 1 and time 2 types of data. Generally it is easier to evaluate "change" between years but using data taken at the same time of year (e.g., June to June).

2.) Cloud condition:
If you have partial cloud cover it creates highly dynamic and variable light intensities and can change the spectral quality of light. So, it is almost impossible to collect good field spectral data under cloudy conditions. Often people "see" the cumulus type clouds but don't see the more wispy and thin cirrus clouds-which actually cause more difficulty in interpretations. Some places almost always have cloud cover, but are important to measure. What can you do? Measure calibration spectra before and after each measurement spectra. This increases time requirements but improves your data quality for the analysis. Compromise on the time of day that field data is collected. It is better to make measurements early in the day (if afternoon clouds are the issue) and assume a correction for the solar angle.

3.) Aerosol, haze, and water vapor:
Aerosols, smoke and water vapor in the atmosphere also deteriorate quality in image data. Often there is nothing you can do about changing the dates for the work, so you are stuck with these conditions. There are no good ways to "calibrate out" the effects of aerosols and smoke. Water vapor can be modeled relatively good in the spectra. Having access to weather station data, micrometeorological data are useful later when analyzing the image data. If you want a more rigorous calibration of these atmospheric properties they can be measured using a "sun photometer," sometimes called a "Regan Radiometer." These are used to measure transmission of direct beam light at different wavelengths through the atmosphere. The measurements are made at multiple sun angles and the data is entered into a calibration program that provides information on optical depth, visibility, O3, water vapor, etc. This would typically be done on the day of the overflight to provide corrections for calibrating surface reflectance in the image data.

4.) Topography:
The interactions between sun angle and topography determines whether the particular slope is in partial shade or is getting reflected light from another slope. It is important to keep this in mind when establishing the sampling design. Some sites only have shade free access at specific times.

5.) Shadows:
To minimize shadows in the field data (shadows cause variability in the overall brightness of the reflected spectra), the best method is to lay out a transect that is in the "principal plane of the sun." That is, it is oriented along a track that is parallel with the direct beam of the sun. If you stand with your back to the sun and orient yourself in the line of least shadows, you will be in this trajectory. You can record your direction of the transect with a compass or with GPS.

Spectral sampling

1.) Spectral Averaging:
Most field spectra are noisy. Consequently, it is best to record several replicates of each sample and save the mean. Most field spectrometers give you the option of selecting the "default" or specified number of replicates to be averaged. These are averages of the "same" measurement-that is, the spectrometer is not moved between measurements so it remains focused on the same object. Typically replicates use 5 or more repeated measures to lower the sample noise. Because it takes some time to repeat these spectral samples, it decreases the total number that can be sampled. So, you need to make a decision about where is your greatest source of error. If you want to map vegetation and you want to sample as many sites as possible, then minimizing this number of replicates may improve your image mapping capability. If however, you want to determine plant stresses (assuming these are small spectral shifts), then you may want to increase the replicate averaging number and get fewer samples (e.g., stressed and no stress conditions).

2.) Sampling intensity:
How many samples of each "image category"? This is the most difficult question to answer. You need enough spectra of the types to believe that at least for the major types, you have sampled the full range of conditions-or a sufficiently broad range that most of the field variance is known. This can be done statistically, but it is rare that you could make measurements of enough samples to show that all the variance is captured in the dataset. Instead, you can make several measurements (e.g, 10 objects of the same type) and then continue to make more measurements until the new spectral measurements fall within the 1+/- standard deviation. If the goal is to identify average conditions or some known condition, this type of model works pretty well.

An efficient way is to do this is a two step field measurement design (if you have repeat access to the site). In this plan, you make measurements of the major classes (or categories) and use these to analyze the image data. Then identify either these classes in new (unsampled) locations or "anomalies," that is, things you didn't expect. Then using a GPS, locate these objects in the field and confirm their identity or conditions.

The types of categories that need to be sampled:

1. Calibration Targets:
Large patches of a spectrally "stable" target that can be used for calibrating the image to surface reflectance. There are empirical and physically based models for calibrating images but you need some "known" reflectance targets to quantify if the calibration is good or not. A dark and light target are desired. The dark target can be a lake or water body (preferably one without algae and sediment). Asphalt makes a good dark target. A bare soil can be used, preferably one with a monotonic (continuous) spectrum and no narrow absorption features (however if it is measured in the field you have some validation if a feature is present). Cement, aluminum can be used as a light target. Using objects in the field has another purpose, this is checking the geographic registration of the location of the material in the image.

2. Dominant types of vegetation and land cover in the image. Spectral measurements of all soil types are desired. Note if the soils are dry or wet at the surface. Major geologic outcrops. It is useful to know the parent material. Spectra of the major classes of litter. Often this might be current or recent litter and "old" litter, since the color often changes. Wood, bark, leaf litter of different species should be measured. The major plant species that form the overstory in the area. Any other materials in the image that are likely to be important to the anlaysis. A city on the edge of the image may not be relevant for the analysis and can be ignored. Here the goal is to let large errors occur in objects that are outside the project interest and minimize errors for critical soils and plants.

3. Control sites:
If looking for effects of specific physical or environmental influences, it is likely that it will be necessary to look at similar nearby areas which can be used as a control site for comparison.

How to measure spectra:

1.) Instrument Position:
The instrument should be set up at right angles to the surface (at "nadir," looking straight down). The instrument should be far enough away to obtain a reasonable average of the materials that characterize the category. That is, if a shrub canopy is being measured, you want to get a representative average of the green foliage, dead foliage, stems, and holes in the canopy that let light penetrate to the ground surface. There is a way to calculate the field of view of the sensor, to determine the height you need that is based on the lens angle and distance away from the surface.

2. ) Avoid Shadows:
Always avoid shadows within the measurement, so orient the instrument toward the sunlight). Orient along the principle plane of the sun to minimize shadows. This will yield the most consistent data set.

4. Reference Calibration Standard:
Measure a white calibration standard every few minutes to prevent instrument drift and recalibrate for changes in sun angle. The NBS standard reflectance material is called "Spectralon" a spun glass material that has very high reflectance across the 400-2500 nm range. It is made by Labsphere, Inc. Calibrate about every 10 min. under good sky conditions, oftener if sky conditions are poor or visibility is low. Be sure panel is level and look for specular reflection (where light is preferentially scattered in a particular direction-it is easy to see if when you look at it some orientation is very bright relative to other orientations. If using the GER 2600, this file is automatically included with all subsequent files referenced to this, i.e. until you take another reference scan. However, on other instruments, e.g. GER 1500, it is important to record this file as a reference calibration standard.

5. Avoid light contamination of samples:
Don't wear bright colored clothing because the light reflects back to the object being measured and into the spectrometer. If you wear a bright red, yellow, green or blue clothing it can change the reflectance of the material being studied. Don't stand where you are shading the object.

6. When using a bucket truck or other platform:
If you work from a bucket truck, the problems are leveling the spectrometer, getting the reflectance panel in place below the spectrometer (and close enough so nothing else if in the FOV), and avoiding sun glint off the metal parts of the truck. Use a black cloth draped around the bucket to prevent glare and reflectance. Remember that cotton cloth is made of plant fibers and is primarily of cellulose, a material that has features in the infrared and one that you want to detect in the field data. So, be cautious and record what is used and what effect it has on the measured spectrum.

7. Optical Field of View (FOV):
Each lens or fiber optic option on the spectrometer has its own field of view, i.e. the size of the area that gets sampled when the instrument is at a specific height. Knowing the actual on the ground sampling size for the field of view with the instrument at various heights will improve the accuracy of sampling, particularly vegetation description, and help to make sampling decisions.

Sampling of bio-physical parameters

1.) What kind of samples to take:
Which types of samples to take, and how many, is dependent on the data required to answer a particular research question, types of analysis to be performed, physical limitations and time constraints. Certain biophysical parameters may be measured on site, e.g. soil color, pH, leaf area index, canopy ht., etc. Generally, however, a small sample of either plant material, ground litter, or soil, is taken in conjunction with the hyperspectral measurement to be analyzed later at a soils or plant laboratory.

2.) Plant or Litter Samples:
Plant or litter samples are taken for a variety of reasons and there is an option of many analyses, typically physical measures, such as biomass wt., or chemical analysis, like leaf nitrogen status. It is relatively easy to measure fresh weight and dry weight (after drying), so that would be the quickest and easiest to do. It is also relatively easy to get total chlorophyll or chlorophyll a and b concentrations. Almost all other chemical measurements are time consuming, and can be expensive when done in mass. It is useful to make a list of the required analyses, and the amount of biomass required by the plant analysis lab for each desired analysis. Also find out from the lab the proper handling procedures for samples, for instance, should samples be kept on ice, frozen, and how long is the maximum allowable time before reaching the lab. Collection of plant samples should be rigorously recorded, including condition of the plant/leaf/tissue at time of collection, e.g. slightly senescent or understory leaf. Plant samples are also sometimes required to perform further hyperspectral analysis upon returning from the field, e.g. using an integrating sphere to measure individual leaf samples.

3.) Soil Samples:
Soil samples are sometimes collected in conjunction with hyperspectral measurements, and are usually sent to a soils lab for standard physical and chemical analyses. It is useful to make a list of the required analyses, and the amount of soil required by the soils lab for each desired analysis. It would be very useful to know the composition of some soil samples if the goal is identifying possible contaminated sites. Also find out from the lab proper handling procedures, for instance, should samples be kept on ice, can they be frozen, and how long is the maximum allowable time before reaching the lab for a particular analysis. Collection of soil samples should be rigorously recorded, including gps geo-referencing and placement in relation to the hyperspectral measurement. On bare soil, before taking the soil sample, take a spectral measurement of the actual sampling site. Generally, a sampling protocol is decided upon in advance which will give the required amount of material for all analyses to be preformed. For instance, it may be decided to sample along a transect, or every ten spectral measurements, or to pool samples from a larger area to get a representative sample. Depth of the soil sampling should also be addressed.

4.) Sampling Regime:
Since it is generally much slower to make the physical measurement, or take enough physical samples, than it is to measure the spectrum, a statistically randomized sampling regime is rarely included as part of the actual hyperspectral fieldwork. The usual method is to make a subset of measurements that have both spectra and biophysical measurement and then extend this relationship through the more quickly measured field spectra. There is a large literature on many types of biophysical measurements. It may be possible to get a rough approximation (at least on a relative basis) by applying an empirical biophysical algorithm or allometric relationship to field or image spectra without such data on the specific site. The validity of this depends on how critical accuracy is to the specific project.

Georeferencing of Spectral Measurements and Imagery:

All field data should be acquired with differentially corrected GPS at 1m (or better) accuracy. This spatial accuracy is easy to obtain at the same time as the field spectra. It is faster and avoids problems if real-time DGPS is used. Be sure to cross-reference the data sets so these data can be related to each other for image analysis.

1.) Coordinate system:
Usually it is a good idea to set the coordinate system parameters to the existing GIS database, map, or other existing spatial referenced data. Important considerations include projection, datum, spheroid, and units, among others. In any case, points acquired by the GPS can easily be converted to other projections. The important thing is to not get confused if using the GPS to locate points on the map. Never change the coordinate system half-way through data collection, or if you do, record this carefully and make the change obvious in the notes to avoid later confusion.

2.) Check satellite signal for data quality:
The strength of the satellite signal during the planned period of the fieldwork can be checked to help in data collection planning. This information is available on-line at: http://www.trimble.com/satview/. It is possible to select the time period, the site position and the mask (the standard value on the Trimble is 15 degrees), and return a graph of PDOP (position dilution of precision) over the planned fieldwork time period. This may be more important if a high level of precision is required. Work breaks may need to be planned in conjunction with periods of poor coverage.

3.) Data logging options:
Generally there are several configurable fields that control the manner in which the GPS data are logged. Keeping the standard ones can be a good option, unless specific requires specify otherwise. Using 60 fixes for averaging position usually gives a good estimate of the point in a reasonable amount of time. It may be possible to reduce this number, if the PDOP is low (i.e. data quality is high), and high precision is not required. Usually, the PDOP mask is set to 6, which is already not an optimal value for data collection. When using the real-time DGPS, check that the radio corrections indicator is turned on, signaling that the differential correction signal is being received.

4.) Data dictionary:
If the characteristics of the environment that is going to be measured is known (.e.g. vegetation types), then a data dictionary can be set up for the data logger. A data dictionary is a customized data entry procedure, allow feature to be easily annotated. The data dictionary can significantly improve the efficiency of certain types of data collection. It is especially easy to download the data afterwards for further processing.

5.) Georeferencing the spectral measurements:
It is particularly important that a clear, accurate, and precise system if used to relate the GPS files and their respective feature points to the spectrometer files. One approach is to name the GPS file the same as the (GER 2600) spectrometer file (e.g. TA1). Individual marked features with the (Trimble) GPS file are then annotated in the field with the file (reading) number of the spectral measurement. For example, the GPS file "TA1" contains seven features, each labeled from 0 to 6 respectively, corresponding to the spectrometer files "TA1.000" through "TA1.006". Another possibility is to keep the suggested file from the data logger (this filename indicates the date, time and a sequential feature number), and rigorously record this information with each measurement.

6.) Georeferencing ground control points for image geo-correction:
It is important to get enough tie-points to geo-reference the image data, both to acquired measurements and other available spatially referenced data. As spatial resolution of the images increases, geo-referencing become more difficult. Often maps and other supporting data have errors larger then the image data. So, good ground truth includes GPS points. Because of the distortions in aerial photography due to roll, pitch and yaw of the plane, it is essential to have tie-points throughout the image. You need objects you can identify in the image and you need to identify the coordinate position. Useful features include road intersections, building corners, edges of fields, or other sharp boundaries. Locations can be compared to existing maps, but maps are often wrong at these high spatial resolutions, so again, it is an iterative procedure to find the "best fit" for the location. You can walk around something (like a clearing or meadow) with the GPS and record the boundary. Because you have measured a polygon, there is only a small location uncertainty that can fit into the appropriate space on the image. You can drive or walk along roads and record vector orientations. You can walk around large trees to record their canopy diameters. Because GPS gives elevation and spatial location, you can compare elevation with a topographic map.

Data transcription, error checking, and compilation of spectral database

In many ways these are the most critical steps in the data collection. Usually, this is done in the lab and rarely in the field. The field spectra need to be transcribed to a database. All data sets need to be compiled and evaluated for errors. Then the spectra must be reviewed (tedious without a special program to do it) and any "bad data" eliminated. Bad data can happen when somehow the spectrometer only records part of a spectrum or there was something wrong with the file that was saved. Or you can't match a "calibration panel reading" with the sample spectrum so it can't be calibrated to reflectance. Or because you have 40 spectra of that material and the sample spectrum is > 3 S.D. from the mean. Often there are errors in data recording. A sample is identified as 100% X but it is actually a mix of X and Y. Often these types of mistakes can be deduced from the entire data set, and in fact you can estimate how pure the unknown sample spectrum is by "mixing" X and Y pure spectra. These steps take concentration and careful checking and the actual processing is somewhat tedious. The major concern here is carefulness.

Preliminary evaluation of data quality

At this point you can determine the statistical differences among the categories in the dataset. If some categories have very large variance, it might be worth additional time in the field to collect more measurements of the material and try to break the category into two or more sub-types that are better characterized. If it is not possible to re-measure, some insight may be obtained by comparing data against field notes. What characterizes data at either end of the variability? Is there some absorption feature that indicates a specific biophysical condition that may inform the image analysis?

Field data is generally noisy, especially at longer wavelengths in the infrared. This is due to several factors, low irriadiance from the sun at wavelengths longer then 1.5 um., low sensitivity of spectrometer sensor materials. This often causes noise around 1050 nm (where silicon based detectors are replaced by another type of detector) and at wavelengths longer then 1.5 um. This is because sensors have low sensitivity in this region. Lastly, plants and wet soil have low reflectance in this region, making the measurements subject to the noise factors above. It is possible to use a smoothing function to eliminate some of the noise in the data but this may eliminate small spectral signals, so it isn't always better to remote noise.

Most standard statistical measures assume independence of data, so it is not necessarily appropriate to use these to compare spectral measurements when the band-to-band variance is highly correlated.

Preliminary analysis

Typically at this stage, we determine if we need to go into the field again to make more spectral measurements. If the image data indicates some major land cover type that was not measured, we will try to collect it in a second trip. This provides a chance to test data analysis by identifying sites that were not visited in the first study and targeting them for verification measurements.

Preliminary analyses are compared against published literature, existing spectral libraries (e.g., JPL, USGS and ISPRA (Italy) have spectral libraries). Information about plant stresses from the literature are also used to confirm interpretations.

Summary Report

It can be particularly useful to summarize the fieldwork campaign in a brief summary report. This can be especially helpful to others who may want to use this data, and preserves the value of the data set over time. This summary should explicity list types of measurements and samples, sampling protocols, file naming conventions, and contain a spreadsheet of the data associated with the spectral measurements. All data resulting from the fieldwork could then be put on one CD with the accompanying report is safekeeping and distribution.

Appendix 2:

Check List of Standard Procedures

1. Collect field spectral data within 2 hrs before and after solar noon. Avoid earlier and later data collections if possible.

2. Collect data under cloud-free sky conditions. Clouds on the horizon are ok but overhead clouds will result in poor or unuseable data. Thin cirrus clouds significantly affect light quality and intensity even though you don't see much with your eyes. Avoid making measurements if cirrus clouds are overhead.

3. Collect GPS data with all sample site data for location information. Make written notes on site condition, percent cover by vegetation type, and note any unusual conditions to aid later interpretation. Note the time and date on the data sheets; cross reference spectra and GPS file names to avoid confusion later. Obtain a site picture for a permanent record of the condition.

4. Orient the entrance on the fiber optics head directly downward (at Nadir). Keep it within + or ­ 10 degrees of perpendicular to the horizontal direction (use a plumb, bubble level or inclinometer to insure this orientation).

5. Record the height the sensor is above the surface being measured (soil or plant canopy). This allows calculation of the field of view. If the spectrometer has more than one foreoptic option, record which one is being used. Be sure to recalibrate if you change the FOV or the integration time (the time the spectrometer takes measurements-equivalent operation to the time the lens is open on a camera).

6. If you have a choice of foreoptics (i.e., the field of view that the sensor admits light into the fiberoptics), then pick one that gives a relatively large FOV on the ground so that there is more spatial averaging. Knowing the FOV at the measurement surface will avoid someone standing in the measurement area or measuring other "extraneous" materials (that is things that are not part of the scene) or shadows. The measurement diameter (at the surface) is equal to the height of the spectrometer above the surface x the FOV of the entrance optics (i.e., the solid angle that it admits light). If the height is in meters and the FOV is in radians, then the diameter of the field of view of the surface being measured will be calculated in meters. If the foreoptics is expressed in degrees, one radian = 57.29degrees, for calculating the appropriate units. Typically the sensors have entrances of milliradians (10-3 m).

7. Collect 5 replicates at the same location that are averaged within the spectrometer for each file saved. This minimizes noise in the data.

8. Collect white panel calibration readings before and after each set of data spectra. Under clear sky conditions near solar noon, these readings can be 10-15 minutes apart. Under less optimal conditions, obtain calibration readings every 5 minutes.

9. Avoid shadows on the panel or within the target being measured. Avoid wearing bright clothing that can change the color of the light being measured. Wear white, black or neutral colors.

10. Collect spectra of reference sites for calibrating the images. Locate spectrally uniform and unchanging (within practical limits) light and dark targets (level ground of bare soil, sand, asphalt or cement, etc.) to use for calibration purposes. At least one of each type are needed. The uniform areas should be at least 3 x 3 pixels in extent (if possible). Collect data from 5 or more distinct locations (e.g., a central point and 4 points in cardinal directions) within each area to provide information on spatial variability within the target site. Collect data about 1 pixel distance (or greater) away from the central point. If the sites are large, collect point location data over several pixels.

11. Locate and collect spectra on as many different bare soil types as possible within the image. These should represent the widest range of soil conditions that are known to be present in the image. Collect 5 or more location points for each of these soil/surface site conditions. Assuming these areas are larger then a pixel, collect data over several pixels. If there is a gradient, orient the data collection across the gradient and take measurement points along a line at 0.5 pixel intervals across the gradient.

12. Collect transect data along the principle plane of the sun. This is done by orienting yourself with your face to the sun and your shadow in a line directly behind you. This minimizes shadow in the data.

13. Identify the most contaminated site(s) in the region and make spectral measurements of the canopies of the most common species (identify them by name). Take at least 3 spectra of points within each plant canopy to get a mean spectrum of the canopy. Take spectra of at least 5 plants of each species in the contaminated area.

14. Locate a similar "uncontaminated" site that has the same soil and plant species as the contaminated site (as much as is practical). Collect the same range of spectra following the procedures on the "contaminated plants and soils" site(s).

15. Locate any potential "intermediate condition" sites and make a similar suite of measurements over these sites.

16. Obtain as many sets of site measurements as possible within the time constraints. Always try to measure things that are "common" within the area and spectra of the "unusual" conditions that have potential to stand out in the imagery. It is impossible to collect "too much." Focus on collecting any unrepresented (in previous measurements) site conditions to expand the database.

Standard Field Spectral Protocol

Check List of Standard Procedures

17. Collect field spectral data within 2 hrs before and after solar noon. Avoid earlier and later data collections if possible.

18. Collect data under cloud-free sky conditions. Clouds on the horizon are ok but overhead clouds will result in poor or unuseable data. Thin cirrus clouds significantly affect light quality and intensity even though you don't see much with your eyes. Avoid making measurements if cirrus clouds are overhead.

19. Collect GPS data with all sample site data for location information. Make written notes on site condition, percent cover by vegetation type, and note any unusual conditions to aid later interpretation. Note the time and date on the data sheets; cross reference spectra and GPS file names to avoid confusion later. Obtain a site picture for a permanent record of the condition.

20. Orient the entrance on the fiber optics head directly downward (at Nadir). Keep it within + or ­ 10 degrees of perpendicular to the horizontal direction (use a plumb, bubble level or inclinometer to insure this orientation).

21. Record the height the sensor is above the surface being measured (soil or plant canopy). This allows calculation of the field of view. If the spectrometer has more than one foreoptic option, record which one is being used. Be sure to recalibrate if you change the FOV or the integration time (the time the spectrometer takes measurements-equivalent operation to the time the lens is open on a camera).

22. If you have a choice of foreoptics (i.e., the field of view that the sensor admits light into the fiberoptics), then pick one that gives a relatively large FOV on the ground so that there is more spatial averaging. Knowing the FOV at the measurement surface will avoid someone standing in the measurement area or measuring other "extraneous" materials (that is things that are not part of the scene) or shadows. The measurement diameter (at the surface) is equal to the height of the spectrometer above the surface x the FOV of the entrance optics (i.e., the solid angle that it admits light). If the height is in meters and the FOV is in radians, then the diameter of the field of view of the surface being measured will be calculated in meters. If the foreoptics is expressed in degrees, one radian = 57.29degrees, for calculating the appropriate units. Typically the sensors have entrances of milliradians (10-3 m).

23. Collect 5 replicates at the same location that are averaged within the spectrometer for each file saved. This minimizes noise in the data.

24. Collect white panel calibration readings before and after each set of data spectra. Under clear sky conditions near solar noon, these readings can be 10-15 minutes apart. Under less optimal conditions, obtain calibration readings every 5 minutes.

25. Avoid shadows on the panel or within the target being measured. Avoid wearing bright clothing that can change the color of the light being measured. Wear white, black or neutral colors.

26. Collect spectra of reference sites for calibrating the images. Locate spectrally uniform and unchanging (within practical limits) light and dark targets (level ground of bare soil, sand, asphalt or cement, etc.) to use for calibration purposes. At least one of each type are needed. The uniform areas should be at least 3 x 3 pixels in extent (if possible). Collect data from 5 or more distinct locations (e.g., a central point and 4 points in cardinal directions) within each area to provide information on spatial variability within the target site. Collect data about 1 pixel distance (or greater) away from the central point. If the sites are large, collect point location data over several pixels.

27. Locate and collect spectra on as many different bare soil types as possible within the image. These should represent the widest range of soil conditions that are known to be present in the image. Collect 5 or more location points for each of these soil/surface site conditions. Assuming these areas are larger then a pixel, collect data over several pixels. If there is a gradient, orient the data collection across the gradient and take measurement points along a line at 0.5 pixel intervals across the gradient.

28. Collect transect data along the principle plane of the sun. This is done by orienting yourself with your face to the sun and your shadow in a line directly behind you. This minimizes shadow in the data.

29. Identify the most contaminated site(s) in the region and make spectral measurements of the canopies of the most common species (identify them by name). Take at least 3 spectra of points within each plant canopy to get a mean spectrum of the canopy. Take spectra of at least 5 plants of each species in the contaminated area.

30. Locate a similar "uncontaminated" site that has the same soil and plant species as the contaminated site (as much as is practical). Collect the same range of spectra following the procedures on the "contaminated plants and soils" site(s).

31. Locate any potential "intermediate condition" sites and make a similar suite of measurements over these sites.

32. Obtain as many sets of site measurements as possible within the time constraints. Always try to measure things that are "common" within the area and spectra of the "unusual" conditions that have potential to stand out in the imagery. It is impossible to collect "too much." Focus on collecting any unrepresented (in previous measurements) site conditions to expand the database.