Quantifying forest disturbance using LiDAR data and time series Landsat images / Syaza Rozali
Tropical rainforest in Malaysia experienced logging activities either legal or illegal since 1960s. Natural regeneration become slower if the permanent production of forest is takes place in inappropriate manner until today. Remote Sensing technology is widely used in forestry application to preserv...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/60955/1/60955.pdf https://ir.uitm.edu.my/id/eprint/60955/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Tropical rainforest in Malaysia experienced logging activities either legal or illegal since 1960s. Natural regeneration become slower if the permanent production of forest is takes place in inappropriate manner until today. Remote Sensing technology is widely used in forestry application to preserve the natural resources. Airborne LiDAR sensor (ALS) is one of the technologies that can be applied together with satellite imagery available such as Landsat and has been proved as an efficient sensor in managing and analyzing deforestation and forest degradation in tropical rainforest. This research aim is to quantify forest disturbance using multi-sensor Remote Sensing in tropical rainforest. The research objectives are to determine forest and non-forest cover from Landsat time series in 2007 until 2018 and Airborne LiDAR sensor data by using vegetation indices (NDVI) and canopy height model (CHM); to identify the forest disturbance detection from Landsat time series by measuring the history period and structural change using BFAST model, and disturbance index (DI); and to quantify the forest disturbance in tropical rainforest using fusion image of spectral and height data by resampling Landsat (30m) into (1m, 15m) resolutions, object-based segmentation technique and classification method using Nearest Neighbor, Random Forest and Support Vector Machine. Two fusion data are tested; 1) SpectralLandsat and 2) SpectralLandsat + HeightALS by Random Forest and Support Vector Machine classification algorithm. The result shows second fusion data having 1 meter Landsat resolution and Airborne LiDAR performed better classification using object-based segmentation and Random Forest classification, about 96% of the overall accuracy with 0.91 kappa index of aggreement. Using the scale of interpretation for kappa statistic shows the very good agreement. Mcnemar‘s test (p-value <0.05) for SpectralLandsat + HeightALS using Random Forest classifier is 0.03. All objectives were achieved successfully and the findings shows that; 1) the higher the resolution of the fusion image, the higher the number of the scale parameter will be used in multi-resolution segmentation; 2) the accuracy of classification was improved when combining LiDAR and Landsat image and 3) quantifying the forest disturbance can be performed using NDVI, CHM and DI information. |
|---|