USGS Spectroscopy Lab

Verification of Remotely Sensed Data

Trude V.V. King and Roger N. Clark
U.S. Geological Survey, MS 964
Box 25046 Federal Center
Denver, CO 80225-0046

Derived from (use the following reference):
Trude V.V. King, Roger N. Clark, Verification of Remotely Sensed Data, in Remote Sensing for Site Characterization (F. Kuehn, T. King, B. Hoerig, D. Pieters, eds.), Springer, Berlin, p 59-61, 2000.


Ground or field checks are an important part of any remote sensing study and are necessary to provide an accurate and useful interpretive product. Field checking is necessary to confirm the validity of spectral, spatial, and morphological interpretations. In general, field checking should be done during all stages of any type of a remote sensing investigation. The methods and magnitude of work necessary to complete the field checking will be dependent on the type of remote sensing data to be verified and the scientific questions to be answered. Remotely sensed data provides an assessment of natural and anthropogenic features as they appear at the time of data acquisition, and possible changes between data acquisition and field checking must be considered.

Verifying historical aerial images can be a difficult, but very important task in a remote sensing study. It is especially important to attempt to authenticate potential links between complementary data sets which were collected at different times. The cultural and natural character of an area may have changed dramatically over time, requiring careful analysis of the data to establish commonality between the modern and historical images. The historical data may contain valuable information about physical or cultural conditions that have been obscured, changed or eliminated, but are crucial to ongoing studies. Establishing linkages between historical and current data is particularly vital when attempting to reconstruct events at a hazardous waste site where physical or environmental modifications have taken place.

Virtual Versus In-Situ Verification

There are two types of verification of remote sensing imagery information: virtual and in situ. Virtual verification can be done by examining the remote sensing data directly: there is sufficient spatial and/or spectral resolution to positively identify objects in the image. In situ verification requires visitation to the area of interest and direct sampling of the environment to verify the remotely sensed information. The following are examples of the application of the methods of verification.

Example 1: red cars. Consider a hypothetical problem of needing to determine the number of red cars in a parking lot using remote sensing data. It is essential to separate cars from other vehicles, as well as distinguish the specific color. If Landsat TM data were the only available data set, with 6 visible to near-IR bands and 30-meter pixels, it would be impossible to resolve car shapes. In addition, a spectral unmixing algorithm to find pixels containing red signatures would be necessary. However, if an analysis was applied, the resulting classification image would not be positive identification of red cars. In addition, the field verification would be needed to confirm that the derived red areas actually contained red cars. In situ field checking would have to be done nearly simultaneously with the data acquisition because the distribution of cars in the parking lot might change. To verify the image information, red cars would have to be accurately located in the parking lot and on the image data to assess the accuracy of the classification. If the classification is shown to be accurate, the classification scheme could be extrapolated to other parking lots to give an indication of the number of red cars in those lots.

Cars could be distinguished from other vehicles on high resolution black and white imagery but the cars could not be determined with 100% accuracy. Subtle differences in the gray scale might allow red cars to be distinguished with some degree of reliability. In-situ verification would need to be done nearly simultaneously with the data acquisition to ensure accuracy in the identification of red cars.

However, color imagery with sufficient spatial resolution to distinguish cars from trucks, and sufficient spectral resolution to discriminate the color red, would allow all red cars in the parking lot be identified with 100% certainty without any in situ field checking. Simple examination of the imagery would allow red cars to be identified and counted much the same as would be possible by being in the parking lot at the time the data were acquired. Such positive identification gained directly from the remotely sensed data is called virtual verification.

Note that there is a distinction between "identification" and "classification" of results from the analysis of the remote sensing imagery. While some objects can be identified directly from the imagery, others can only be inferred to some level of confidence (classification) that will need in situ field checking. Information gained from the analysis of remotely sensed data increases with increasing spatial resolution and/or spectral resolution and, in general, is maximized when both high spectral and high spatial resolution are used together (Figure 1). Although, in some instances only high spatial or spectral resolution is needed. For example, identifying cars requires high spatial resolution, but only poor spectral resolution. Black and white imagery would suffice to identify cars; a simple color photo could be used to determine its color.

Figure 1 caption. The information content of remote sensing data is a function of spectral range, sampling and resolution and spatial resolution. Although narrow sampling intervals, high spectral resolution and a wide spectral range improve the information content in remotely sensed data, similalrily so does high spatial resolution. For optimum science information, ideally both the spectral and spatial parameters would be maximized.

Example 2: Minerals in Soils. If the problem is to identify the presence of a specific mineral, for example calcite, in a large areal exposure of soil, the following should be considered. The use of black and white imagery would provide little chance of locating calcite. Extensive field sampling, including returning samples to the laboratory for analysis, such as X-ray diffraction, would be needed to derive a correlation of ground albedo (image gray level) with calcite occurrence. However, such a classification is not a positive identification as the possibility exists that materials other than calcite could produce a similar response.

Using imaging spectroscopy data, with suitable spectral resolution, it is possible to identify specific minerals in soils, such as calcite, based on the wavelength position and shape of characteristic absorption features (see the online spectral library for example calcite spectra). The detection of unique calcite spectral absorption features allows the positive identification of the mineral and the capability to map its distribution, based on the limits of the spatial resolution of the instrument. In this case, there is no need for in-situ field checking because the spectra are of sufficient resolution to be certain of their identification. The derived calcite maps can be verified by examining spectra from the imaging spectrometer data, a form of virtual field verification.

Increasing the spatial resolution of the instrument would not increase the likelihood of mapping the presence of calcite in the soils, unless the spatial resolution approached that of the grain size of the individual calcite grains. If the spatial resolution approached that of a microscope, individual crystal morphologies could be resolved and the areal distribution of a specific mineral could potentially be mapped using this criteria. Consequently, spectral resolution is more important than spatial resolution for identifying specific mineral composition.

Verification of Vegetation Data

Verification of vegetation spectra is challenging whether the in-situ or virtual approach is used. During the verification process, the spectral signatures of specific plants or plant communities are correlated with the spectral signatures in the remotely sensed data. The absorption features of interest result from the internal cellular structure and chlorophyll, ligand, and water content of the vegetation. Thus, spectra used in comparisons should be from plants having analogous growth cycles and environmental conditions as the remotely sensed plants. In-situ verification should be completed as soon as possible after remote data acquisition, minimizing the spectrally detectable natural chemical and climatic responses resulting from changes in environmental conditions and natural growth cycle. See Kronberg (1985) and King et al. in Sect 6.7 for further discussions.

The problem of viewing aspect adds to the complexity of verification of remotely sensed vegetation information. Remotely sensed data is collected looking downward, similar to what an individual can see from an airplane. The overhead viewing position influences the proportions of leafy green material, bark, and stems seen in individual vegetation species. Consequently, in forested areas the remotely sensed data measures the canopy characteristics, rather than characteristics of individual trees and plants, some of which may grow beneath the canopy.

Logistically, verifying the composition or structure of the vegetation canopy is difficult because of cost or access issues. In-situ verification measurements, helicopter based for example, are desirable, but in most instances they are extremely expensive and often prohibited by access issues. Consequently, verification must be done by comparing weighted spectra of individual components- varying proportions of leaf, stems and bark- to the remote canopy observations.

High spatial resolution data, color aerial photography for example, can sometimes resolve the crown shapes of trees which can be used to identify specific species. Despain (1990) used crown shapes in color aerial photographs to map trees in Yellowstone National Park, but bushes and grasses were too small to resolve the structural detail related to species. More recently, Kokaly et. al (1998) used AVIRIS data to define tree, shrub, and grass species based on their spectral signatures, although crown shapes were not resolved.

Regardless of the type of remote sensing data, virtual or in-situ verification will improve the accuracy and usefulness of the final data product.


Despain, D.G., 1990, Yellowstone Vegetation, Consequences of Environment and Historu in a Natural Setting, Roberts Rinehart Publishing, 239 p.

Kokaly, R.F., R.N. Clark, and K.E. Livo, 1987, Mapping the biology and mineralogy of Yellowstone National Park using imaging spectroscopy. In, Summaries of 7th Annual JPL Airborne Earth Science Workshop, R.O. Green, ED., JPL Pub 97-21, vol 1.

Kronenberg, P, 1985, Fernerkundung der Erde. Enkee , Stuttgart, 394p.

U.S. Geological Survey, a bureau of the U.S. Department of the Interior
This page URL=
This page is maintained by:
Dr. Roger N. Clark
Last modified May 24, 2000.