Project Background
Discussion of Uncertainty
Sources & Acknowlegements

source: http://www.salmonriver.org/




Uncertainty in the data

Some sources of error and uncertainty lie in the data sets that were used in this project. The data that we obtained was from 2 different sources: the 1966 data was from the Government of Canada Land Inventory and the 1986 and 2001 data was from Land Surveys done by the Township of Langley. The 1966 data was in the form of clump data, and was projected in NAD27. For this reason the 1966 data set needed to be converted into NAD83 UTM Zone 10N in order to match the 1986 and 2001 data set projections. Although not much difference was observed, there was a slight difference and a spatial shift in the 1966 clump data between the original NAD27 projection and the new NAD83 UTM Zone 10N projection. Due to the fact that the 1966 data set was clump data, there was far less detail in the polygons themselves when compared to the 1986 and 2001 data sets which may have caused some uncertainty when compared to the land use change from year to year. Although the 1966 clump data is of good quality, coming from a reputable source (Government of Canada Land Inventory), its accuracy is limited by its age, as it was published in 1966. However, this error and uncertainty did not affect the project to an extent where it is wholly inaccurate.


Uncertainty from data manipulation

A major source of error and uncertainty lies in the data manipulation. In order to create constant land use codes that could be compared throughout the 3 annual data sets the 2001 land use codes were used a “base” in which the other 2 data sets were reclassified to match. In 2001 there were 7 land use classes identified, and if land use classes from the 1966 and 1986 data did not correspond to the 7 “base” classes, they were subjectively reclassified. As this data was subjectively reclassified, there is possible error in our interpretation of the original land use codes to the new land use codes. This subjectivity could be a source of potential error if the land use codes were reclassified incorrectly and had an impact on the outcome of our comparisons between Agricultural land use and other land use types and the conversion or reversion of these land uses between the 3 years of data that were compared. However, the 1966 clump data only had 4 land use classes that were constant with the 1986 and 2001 data and did not contain any information that would have allowed a reclassification to commercial, government and institutional, or transportation and communication.

Error Matrices

The error matrix calculations were also a source of error and uncertainty. Although these error matrices were used to determine the accuracy of data when comparing the land uses between 1966-1986 and 1986-2001, some uncertainty still remains. As the 1966 data only contained 4 land use types we could not use the crosstab query to compare any more than 4 land use types with the 1986 data. For this reason we only compared the first 4 land use types in the 1986 data, and disregarded the last 3 land use types. By leaving out the last 3 land use types there may be slight uncertainty within the comparison, however, these data do not directly affect the error matrix comparisons due to the fact that the land use types did not change from Agriculture to “other” land use types, a change from “other” land use types to Agriculture or Agricultural land use remaining constant.


Despite early issues with the data, particularly the lack of continuity between data sets, we derived good results in the error matrices and land use comparisons. As shown in the Results section the calculated land use change from other to agriculture, agriculture to other, and agriculture constant between the comparable years (Table 3) in the error matrices are represented visually and agree with the layouts (land use changes between 1966-1986 and between 1986-2001) which provides greater confidence in the quality of the results. The questions set out in the introduction have been addressed with room for further analysis, should more or better data become available.