Remote sensing for statistics
Satellite images provide very useful information to produce statistics on topics closely related to the territory, such as agriculture, forestry or land cover in general. The first large project to apply Landsat 1 images for statistics was LACIE (Large Area Crop Inventory Experiment), run by NASA, NOAA and the USDA in 1974–77.[1][2] Many other application projects on crop area estimation have followed, including the Italian AGRIT project and the MARS project of the Joint Research Centre (JRC) of the European Commission. [3] Forest area and deforestation estimation have also been a frequent target of remote sensing projects [4][5], the same as land cover and land use[6]
Accuracy assessment and area estimation
Before producing statistical estimates, satellite images undergo several operations. Some of them can be considered pre-treatment phases, including radiometric calibration and orthorectification, so that the images can be displayed in a Geographic Information System (GIS). Further steps produce layers that are conceptually closer to the statistical variables we want to estimate. If our target is estimating the area of single crops, such as wheat, for example, we need classified images in which wheat is one of the classes in the legend. The most traditional image classification algorithms work pixel by pixel. Image segmentation for object-based image analysis (OBIA) is essential in other image analysis fields, but is less critical for agricultural and environmental monitoring.
Ground truth or reference data to train and validate image classification require a field survey if we are targeting annual crops or individual forest species, but may be substituted by photointerpretation if we look at wider classes that can be reliably identified on aerial photos or satellite images. It is relevant to highlight that probabilistic sampling is not critical for the selection of training pixels for image classification, but it is necessary for accuracy assessment of the classified images and area estimation. [7][8][9] Additional care is recommended to ensure that training and validation datasets are not spatially correlated. [10]
We suppose now that we have classified images or a land cover map produced by visual photo-interpretation, with a legend of mapped classes that suits our purpose, taking again the example of wheat. The straightforward approach is counting the number of pixels classified as wheat and multiplying by the area of each pixel. Many authors have noticed that this estimator is generally biased because commission and omission errors in a confusion matrix do not compensate each other [11][12][13]
The main strength of classified satellite images or other indicators computed on satellite images is providing cheap information on the whole target area or most of it. This information usually has a good correlation with the target variable (ground truth) that is usually expensive to observe in an unbiased and accurate way. Therefore it can be observed on a probabilistic sample selected on an area sampling frame. Traditional survey methodology provides different methods to combine accurate information on a sample with less accurate, but exhaustive, data for a covariable or proxy that is cheaper to collect. For agricultural statistics, field surveys are usually required, while photo-interpretation may be better for land cover classes that can be reliably identified on aerial photographs or high resolution satellite images. Additional uncertainty can appear because of imperfect reference data (ground truth or similar). [14][15]
Some options are: ratio estimator, regression estimator [16], calibration estimators [17] and small area estimators [6]
If we target other variables, such as crop yield or leaf area, we may need different indicators to be computed from images, such as the NDVI, a good proxy to chlorophyll activity. [18]
References
- ↑ Houston, A.H. "Use of satellite data in agricultural surveys". Communications in Statistics. Theory and Methods (23): 2857–2880.
- ↑ Allen, J.D. "A Look at the Remote Sensing Applications Program of the National Agricultural Statistics Service". Journal of Official Statistics. 6 (4): 393–409.
- ↑ Taylor, J (1997). Regional Crop Inventories in Europe Assisted by Remote Sensing: 1988-1993. Synthesis Report. Luxembourg: Office for Publications of the EC. Search this book on
- ↑ Foody, G.M. (1994). "Estimation of tropical forest extent and regenerative stage using remotely sensed data". Journal of Biogeography. 21 (3): 223–244. Bibcode:1994JBiog..21..223F. doi:10.2307/2845527. JSTOR 2845527.
- ↑ Achard, F (2002). "Determination of deforestation rates of the world's humid tropical forests". Science. 297 (5583): 999–1002. Bibcode:2002Sci...297..999A. doi:10.1126/science.1070656. PMID 12169731.
- ↑ 6.0 6.1 Ambrosio Flores, L (2000). "Land cover estimation in small areas using ground survey and remote sensing". Remote Sensing of the Environment. 74 (2): 240–248. Bibcode:2000RSEnv..74..240F. doi:10.1016/S0034-4257(00)00114-0.
- ↑ Green, Russell G. Congalton, Kass (2019-01-25). Assessing the Accuracy of Remotely Sensed Data: Principles and Practices, Third Edition (3 ed.). Boca Raton: CRC Press. doi:10.1201/9780429052729. ISBN 978-0-429-05272-9. Search this book on
- ↑ Stehman, S. (2013). "Estimating Area from an Accuracy Assessment Error Matrix". Remote sensing of environment (132): 202–211.
- ↑ Stehman, S. (2019). "Key issues in rigorous accuracy assessment of land cover products". Remote sensing of environment (231).
- ↑ Zhen, Z (2013). "Impact of training and validation sample selection on classification accuracy and accuracy assessment when using reference polygons in object-based classification". International Journal of Remote Sensing. 34 (19): 6914–6930.
- ↑ Czaplewski, R.L. "Misclassification bias in areal estimates". Photogrammetric Engineering and Remote Sensing (39): 189–192.
- ↑ Bauer, M.E. (1978). "Area estimation of crops by digital analysis of Landsat data". Photogrammetric Engineering and Remote Sensing (44): 1033–1043.
- ↑ Olofsson, P. (2014). "Good practices for estimating area and assessing accuracy of land change". Remote Sensing of Environment. 148 (148): 42–57. Bibcode:2014RSEnv.148...42O. doi:10.1016/j.rse.2014.02.015.
- ↑ Mcroberts, R (2018). "The effects of imperfect reference data on remote sensing-assisted estimators of land cover class proportions". ISPRS Journal of Photogrammetry and Remote Sensing. (142): 292–300.
- ↑ Foody, G.M. (2010). "Assessing the accuracy of land cover change with imperfect ground reference data". Remote sensing of environment (114): 2271–2285.
- ↑ Sannier, C (2014). "Using the regression estimator with landsat data to estimate proportion forest cover and net proportion deforestation in gabon". Remote Sensing of Environment (151): 138–148.
- ↑ Gallego, F.J. (2004). "Remote sensing and land cover area estimation". International Journal of Remote Sensing. 25 (5): 3019–3047. Bibcode:2004IJRS...25.3019G. doi:10.1080/01431160310001619607.
- ↑ Carfagna, E. (2005). "Using remote sensing for agricultural statistics". International Statistical Review. 73 (3): 389–404. doi:10.1111/j.1751-5823.2005.tb00155.x.
This article "Remote sensing for statistics" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:Remote sensing for statistics. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
