Research Report 3.12
Evaluation of High Resolution Airborne Imagery and Global Positioning Systems for Monitoring Changes in Agroecosystems
A.J.VandenBygaart, M.D. Wood and B.G.A. Hulshof
Department of Land Resource Science, University of Guelph,
Guelph, Ontario, N1G 2W1
COESA Report No.: RES/MON-012/97
Appendices 1, 2, 3
Wood, M.D.1, VandenBygaart, A.J.1, Shepherd, P.2, Protz, R.1, and Hulshof, B1. 1996. Integration of high resolution GPS and CASI Imagery for Agricultural Soil Landscape Studies. 8th Inter. Conference on Geomatics. May, Ottawa.
1 Department of Land Resource Science, University of Guelph, Guelph Ontario.
The analysis of high resolution (~ 1 m) digital airborne imagery provides an excellent opportunity to develop the techniques and methodologies for the use of the next generation of high resolution satellite imagery. Integration of this imagery with accurate field scale Digital Elevation Models (DEM) derived from Differential Global Positioning Systems (DGPS), affords an opportunity to study the effect of small scale topographic changes and tillage practices on the reflectance characteristics of an individual agricultural field.
In the spring of 1995, a Compact Airborne Spectrographic Imager (CASI) collected imagery in the visible and IR regions over six study sites throughout southern Ontario. The sites exhibited both conventional and conservation tillage practices. Reflectance from the conservation till (no till) sites varied considerably less than their conventional counterparts, showing little effect of topography on reflectance. The conventional tillage plots exhibited a response to topography, with high slope areas showing a higher return in the three optical wavelengths while foot slope regions tended to return low reflectance values.
Results and Discussion
Of the six sites, two were chosen for analysis. To investigate the effect of tillage patterns on spectral reflectance, both conventional till and conservational tillage were represented in the sites. The total change in elevation over each of the sites was 2 to 3 meters with relatively simple slopes. It was evident that two primary factors were influencing the reflectance of each of the sites, those being the effect of topography and tillage effects.
Tillage effects on reflectance
The mean, maximum and minimum reflectance values for each of the six CASI bands are shown in Table 1, and illustrate the effect of tillage patterns on reflectance. The conservational tillage (no till) showed less variation in reflectance for all six bands. This is in response to the homogenizing effect of the stubble or crop residue left on the field after harvest. The amount of crop residue found on the field is highly variable depending on the crop type and the specific conservation tillage practice. The movement of the crop residue from the upper slope to the lower slope tends to be slow and dependant on intense rain and degree of slope.
The no till study site, exhibiting only a 3 m total change in elevation, has a very gentle slope and likely very little movement of crop residue occurs down slope. This results in a fairly uniform distribution of stubble over the entire field which could account for the observed lack of reflectance variation of the no till site. In order to investigate the effect of topography on a no till site, a much more complex slope and varied terrain would be needed.
Hulshof, B.1,Protz,R.1, Wood,M.1, and Fischer,J.2 1997. Identification of agricultural field size and boundaries from Landsat TM data in southern Ontario. International Symposium in the Era of Radarsat 1997. May, Ottawa.
1 Department of Land Resource Science,
2 Provincial Remote Sensing Office, Ontario Ministry of Natural
Providing accurate information derived from satellite data on agroecosystems requires ground truth cognizant of spatial and temporal variations that occur on a regional scale. To better understand the interactions between satellite data and agroecosystems training areas need to be identified. It is in this context that all large agricultural fields (those greater than 14.5 hectares in size) were identified in southern Ontario from Landsat TM data.
To facilitate the identification of large fields from Landsat TM data a rigid unsupervised classification procedure was developed. The classification successfully identified large fields. The spatial resolution of Landsat TM data and natural variability in crops due to soil, topography, nutrient status and available moisture were found to be limiting factors in the identification of large fields.
It was established that the most effective way to locate fields was through the identification of homogeneous areas. The method selected for classifying a fields in southern Ontario was through the use of unsupervised classification using the ISODATA algorithm. The classification methodology developed proved to be successful, identifying most fields larger than 14.5 hectares. The greatest number of fields identified occurred in the 14.5 to 25 hectare class. The optimal time to acquire imagery for the classification of large fields is August.
However, the 25 meter spatial resolution of Landsat TM and natural variability found within agricultural fields proved to be the major limiting factors in the classification. The size of some fields was overestimated due to the difficulties in detecting narrow fence lines or single furrow boundaries. Crops located on smaller fields (i.e. tobacco) will not be represented in the sampling scheme because the 25 meter pixel resolution was too coarse to detect these small fields. Higher resolution imagery is needed improve the accuracy with which large fields are identified. Fields with a broad range of natural variability and thus variations in spectral reflectance values were difficult to classify as one field.
Hulshof, B.1, Protz, R.1, and Allen, B.2 1997. Objective training site selection for regional scale satellite data validation. International Symposium in the Era of Radarsat 1997. May, Ottawa.
1 Department of Land Resource Science, University of Guelph
2 Department of Mathematics and Statistics, University of Guelph
The success of agriculture is dependent on many variables such as crop health, soil nutrient status, soil moisture, and soil organic matter. The aim of this paper is to account for spatial and temporal variability of these factors when selecting training sites for calibration of remotely sensed data in southern Ontario. Many methods used for the selection of training sites are subjective and consequently may reduce the classification accuracy with non-representative, biased training sites.
A stratified random sampling scheme was employed to select agricultural fields larger than 14.5 hectares to be used for calibration of satellite data. Three to four fields were located in clusters at road intersections to improve ground truthing efficiency by up to 70 percent. The number of training clusters per county ranged from 18 in Essex county to 78 in Middlesex. It was difficult to satisfy the dual objectives of selecting training areas representative of spatial variations in agriculture and creating a sampling design that is not prohibitively costly.
A total of 1393 clusters with three or more fields larger than 14.5 hectares at a road intersection were identified as candidate training areas for agroecosystem monitoring in southern Ontario. Sampling in clusters of three to four fields reduced the number of locations to be visited for ground truth by up to 70 percent. The number of strata in each county ranged from a minimum of two in Niagara and Essex to a maximum of nine in Middlesex county. The number of training areas selected in each county ranged from 18 in Essex to 78 in Middlesex. Having 78 training areas in one county decreased the efficiency of the sampling design but may be necessary to represent variations in agriculture due to physiography. Counties such as Kent with only 3 strata and 23 training areas illustrate a more uniform landscape and thus a simpler sampling situation. It was difficult to satisfy the dual objectives of selecting training areas representative of spatial variations in agriculture and to create an efficient sampling design.
Last Updated: May 17, 2011 09:50:41 AM