Geodetski Vestnik (Jan 2010)
Uporaba strojnega učenja za določitev poplavljenih območij – primer poplav v Selški dolini leta 2007 : Application of data mining for determination of flooded areas – Selška valley 2007 floods case study
Abstract
V prispevku je obravnavana uporabnost postopkov strojnega učenja pri ugotavljanju poplavljenih območij v zgornjem delu Selške doline, ki jo je 18. 9. 2007 prizadelo hudourniško deževje. Hitro prepoznavanje poplavljenih območij je ključnega pomena za učinkovito reševanje in povračilo škode od zavarovalnic. Pri tem so zelo uporabni satelitskiposnetki, saj omogočajo hitro določitev poplavljenih območij, tudi če so prizadeti zelo veliki predeli. Za prepoznavanje poplavljenih območij v Selški dolini so bile uporabljene tehnike strojnega učenja zrazličnimi vhodnimi podatki: satelitski posnetek SPOT (multispektralni in pankromatski), indeks NDVI, relief in njegovi izdelki (nadmorska višina, naklon, ukrivljenost), oddaljenost od vodotokov in raba tal. Učni vzorci, ki so bili uporabljeni za oblikovanje modela klasifikacije, so vsebovali 400, 255 oziroma49 vzorčnih točk. : This paper discusses the usefulness of machinelearning procedures for determining flooded areas in the upper part of the Selška valley. The area was affected by torrential rains on 18.9.2007. Rapid identification of flooded areas is essential for effective implementation of rescue operations and damage assessments. In this case, satellite images are very useful because they enable quick identification of flooded areas even in very large areas. To determine the flooded areas, machine learning techniques were applied to different input data. SPOT satellite image (multispectral and panchromatic), NDVI index, relief and its derivatives (altitude, slope, curvature), distance from rivers and land use were used. The learning samples consisted of 400, 255 and 49 sample points, which were used to build three different classification models.
