Järvselja õppe- ja katsemetskonna tormikahjude hindamine fotogrammmeetriliste meetoditega
Laen...
Kuupäev
2022
Kättesaadav alates
09.09.2022
Autorid
Rahu, Oscar
Siim, Karmo
Ajakirja pealkiri
Ajakirja ISSN
Köite pealkiri
Kirjastaja
Eesti Maaülikool
Abstrakt
Uurimistöö eesmärgiks oli tuvastada, kasutades mehitamata õhusõidukiga tehtud
aerofotosid, Järvselja metskonnas tormikahjustuste ulatust ning hinnata ortofotomosaiikide piksli suuruse mõju hinnangu kvaliteedile. Selleks jäädvustati mehitamata
õhusõidukitega 16 km2
ala. Kogutud aerofotodest koostati ortofotomosaiigid programmis
Pix4D ja DroneDeploy. Drooni eBee X ortofotomosaiigi täpsuseks saadi 3,6 cm/px ja
mehitamata õhusõiduki DJI mosaiigi täpsuseks 20 cm/px. Valminud ortofotomosaiikide
põhjal hinnati uuritud alade tormikahjustuste ulatust ning kattuvate alade puhul määrati
pikslisuuruse mõju. Kahe mehitamata õhusõiduki ortofotomosaiikide kattuvusalade
suuruseks oli 89,59 ha. Lisaks uuriti võrdluseks Sentinel-2 satelliitpilte. Satelliitpiltide
piksli suurus oli 10 m/px.
Autorid kasutasid hinnangu andmiseks ArcGIS Pro programmi, kus kasutades
digiteerimise meetodit, jaotati tuulemurdudega alad suuruse järgi viide klassi. Mehitamata
õhusõidukiga eBee X mõõdeti ~1069 ha suurune ala, millelt protsentuaalselt tuvastati
kahjusid järgnevalt: klassis „1 puu“ (0,21%), „2–5 puud“ (0,53%), „6–10 puud“ (0,25%),
„11–30 puud“ (0,39%), „31 enam puud“ (1,94%) ehk kogu ala oli kahjustatud 3,32%
ulatuses. DJI drooniga mõõdeti ~665 ha, millelt tuvastati kahjusid järgnevalt; „1 puu“
(0,17%), „2–5 puud“ (0,27%), „6–10 puud“ (0,14%), „11–30 puud“ (0,30%), „31 enam
puud“ (1,87%) ehk summaarselt oli kahjustatud 2,75% mõõdetud aladest. Kattuvatelt
aladelt leiti, et rohkem kahjusid on võimalik tuvastada eBee X ortofotomosaiikidelt, mis
olid üle viie korra parema lahutusvõimega kui DJI mosaiik. Kokku leiti mosaiikidelt 53,74 ha ulatuses kahjustusi. Sentinel-2 piltide klassifitseerimiseks kasutati kahte erinevat
meetodit. Esimese meetodi puhul võrreldi automaatselt kahte (enne ja pärast tormi
jäädvustatud) satelliitpilti, mille alusel hinnati kahjustuste suuruseks 24,57 ha. Teise
meetodi puhul õpetati programmi näidisalade põhjal tuvastama kahjustunud
tuulemurrualasid ja saadi tulemuseks, et kahjustatud on 173,02 ha metsa.
Uurimistöö käigus selgus, et kõige lihtsam on tuulemurde tuvastada parima
lahutusvõimega ortofotomosaiikidelt, samas 20 cm piksli suuruse juures oli veel võimalik
tuvastada üksikuid murdunud puid. Seevastu satelliitpiltide kasutamisel ei saadud
usaldusväärseid klassifitseerimise andmeid, leides palju valepositiivseid tulemusi aladelt,
mis ei olnud kahjustatud.
The objective of the thesis project was to examine the extent of the damages caused by the storm in the Järvselja forest district, as well as to assess how the size of the pixels in photos change the quality of the examination. Drones were used to measure a 16 km2 area. Using aerophotos, a collection of orthomosaics were developed in Pix4D and DroneDeploy. For the orthomosaic developed using eBee X, the final accuracy was 3.6 cm/px, whereas for the DJI model, it was 20 cm/px. The mosaics were assessed separately as well as compared using overlapping areas. The total acreage of the two orthomosaics was 89.59 hectares. Additionally, the researchers examined the Sentinel-2 satellite data. The pixel size of the satellite images was 10 m/px. For the final assessment, the authors used ArcGIS Pro. Using digitization, the windfalls were classified into five groups. Using eBee X, a ~1069 hectar area was measured, from which the damages were categorized percentage wise: „1 tree“ (0.21%), „2-5 trees“ (0.53%), „6-10 trees“ (0.25%), „11-30 trees“ (0.39%), „31 or more trees“ (1.94%), equating to 3.32% of damaged land. Using DJI, a ~665 hectar area was measured, classifying the damages as follows: „1 tree“ (0.17%), „2-5 trees“ (0.27%), „6-10 trees“ (0.14%), „11-30 trees“ (0.30%), „31 or more trees“ (1.87%), totalling to 2.75%. From the overlapping areas, it was determined that a more precise extent of damages can be observed from the orthomosaics developed using the eBee X, due to its superior image resolution (over five times greater than DJI). A total of 53.74 hectares of damages were found from the mosaics. Two methodologies were used to classify the Sentinel-2 images. Initially, the program automatically compared two satellite images, from which it determined the total forest damages amounting to 24.57 hectares. As for the other method, the program was taught to detect windfall based on given examples, from which it found 173.02 hectares of damaged forest. From the thesis project, it was determined that the most efficient way to detect windfall, is to use high resolution orthomosaics. However, at 20 cm/px resolution, it is possible to determine singular fallen trees. This methodology is superior to using satellite images, as those did not provide sufficient classification data, due to its high false positive rate from areas, which were not damaged.
The objective of the thesis project was to examine the extent of the damages caused by the storm in the Järvselja forest district, as well as to assess how the size of the pixels in photos change the quality of the examination. Drones were used to measure a 16 km2 area. Using aerophotos, a collection of orthomosaics were developed in Pix4D and DroneDeploy. For the orthomosaic developed using eBee X, the final accuracy was 3.6 cm/px, whereas for the DJI model, it was 20 cm/px. The mosaics were assessed separately as well as compared using overlapping areas. The total acreage of the two orthomosaics was 89.59 hectares. Additionally, the researchers examined the Sentinel-2 satellite data. The pixel size of the satellite images was 10 m/px. For the final assessment, the authors used ArcGIS Pro. Using digitization, the windfalls were classified into five groups. Using eBee X, a ~1069 hectar area was measured, from which the damages were categorized percentage wise: „1 tree“ (0.21%), „2-5 trees“ (0.53%), „6-10 trees“ (0.25%), „11-30 trees“ (0.39%), „31 or more trees“ (1.94%), equating to 3.32% of damaged land. Using DJI, a ~665 hectar area was measured, classifying the damages as follows: „1 tree“ (0.17%), „2-5 trees“ (0.27%), „6-10 trees“ (0.14%), „11-30 trees“ (0.30%), „31 or more trees“ (1.87%), totalling to 2.75%. From the overlapping areas, it was determined that a more precise extent of damages can be observed from the orthomosaics developed using the eBee X, due to its superior image resolution (over five times greater than DJI). A total of 53.74 hectares of damages were found from the mosaics. Two methodologies were used to classify the Sentinel-2 images. Initially, the program automatically compared two satellite images, from which it determined the total forest damages amounting to 24.57 hectares. As for the other method, the program was taught to detect windfall based on given examples, from which it found 173.02 hectares of damaged forest. From the thesis project, it was determined that the most efficient way to detect windfall, is to use high resolution orthomosaics. However, at 20 cm/px resolution, it is possible to determine singular fallen trees. This methodology is superior to using satellite images, as those did not provide sufficient classification data, due to its high false positive rate from areas, which were not damaged.
Kirjeldus
Magistritöö
Geodeesia, kinnisvara- ja maakorralduse õppekaval
Märksõnad
magistritööd, tuulemurd, mehitamata õhusõiduk, ortofotomosaiik, fotogramm-meetria, Sentinel-2, klassifitseerimine, Roheline Ülikool (töö toetab EMÜ Rohelise Ülikooli põhimõtteid), elurikkus, pärandkooslused, pärandmaastikud, maastikukaitse