نوع مقاله : مقاله پژوهشی
نویسندگان
1 مربی گروه سیستم های فضایی، پژوهشکده سامانه های فضانوردی، پژوهشگاه هوافضا
2 کارشناس ارشدسنجش ازدور، گروه مهندسی نقشه برداری، دانشگاه صنعتی خواجه نصیرالدین طوسی
چکیده
مدل نگارگر انبوهی تاجپوشش جنگلی (FCD) یکی از روش های مرسوم سنجش از دوری در ارزیابی تراکم جنگل است. دو ضعف بزرگ این مدل عدم امکان پیاده سازی برای سنجنده های فاقد باند حرارتی و دقت مکانی پایین نتایج حاصل از آن در جنگل های جلگه ای بخصوص در محدوده های مطالعاتی بزرگ مقیاس می باشد. هدف اصلی این پژوهش گسترش و بهبود این مدل بر اساس دو ایده می باشد. در ایده اول شاخص حرارت مورد استفاده در این مدل را با دو شاخص NLI و GNDVI، که بر اساس یک آزمون آماری مناسب تشخیص داده شدند، جایگزین شدند. بدین ترتیب اجرای این مدل برای سنجنده های فاقد باند حرارتی فراهم ممکن شد. در ایده دوم، به منظور غلبه بر چالش متغیر بودن شرایط جوی و دمایی یک منطقه با وسعت بزرگ، یک مرحله حد آستانه گذاری محلی به این مدل افزوده شد تا دقت مدل برای مطالعات بزرگ مقیاس بهبود یابد. به منظور بررسی عملکرد مدل پیشنهادی، از داده های لندست 5 مربوط به سال 1366 و داده های لندست 8 مربوط به سال 1396 برای محاسبه تراکم تاج پوشش جنگل های هیرکانی شمال کشور استفاده شد. انتخاب این منطقه و این داده ها ضمن روبرو کردن مدل پیشنهادی با یک منطقه مطالعاتی وسیع، امکان بررسی تغییرات مساحت تاج-پوشش جنگل های هیرکانی در طی 30 سال گذشته ( از سال 1366 تا 1396) را فراهم کرد. نتایج نشان داد دقت کلی و ضریب کاپای مدل FCD بهبودیافته نسبت به مدل FCD اولیه به ترتیب حدود 10 و 14 درصد افزایش پیدا کرد. همچنین نتایج این بررسی نشان داد مساحت جنگل های هیرکانی از سال 1366 تا سال 1396 در ارتفاعات پایین دست کاهش چشمگیری داشته است.
کلیدواژهها
عنوان مقاله [English]
Introducing a new local FCD model based on local adaptive thresholdingwith the aim of estimating forest canopy density over large areas
نویسندگان [English]
- Masoud Taefi Feijani 1
- Saeed Azadnejad 2
1 Instructor, Research scientist, Aerospace research institute space systems research group, Tehran, Iran
2 MSc Student in Remote Sensing, Geomatics Engineering Faculty, K.N. Toosi University of Technology, Tehran, Iran
چکیده [English]
Extended Abstract
1- Introduction
The present study primarily sought to present a new FCD model to eliminate two limitations of the initial FCD model.These limitations included the fact thatimplementing the initial FCD model for sensors without a thermal bandwas not possible, sincethe model took advantage of a combination of shadow index and thermal index to detect black soil and calculate advanced shadow index.To overcome this limitation, we replaced thermal index with NLI and GNDVI indices, and combined shadow, BI, NLI and GNDVI indices to detect black soil and calculate advanced shadow index.Defining a global threshold for thermalindex used fordetectingblack soil was the next limitation of initial FCD model.Due to variations in regional climate and temperature, selecting a global threshold for the whole scene does not seem logical.Thus, a local thresholding process was used to define a threshold level for BI, NLI and GNDVI indices.In this regard, the study sitewas divided into 14 sections and an appropriate threshold wasselected for each section.A digital elevation model was also used to define a specific threshold level for forests in flat areasand elevated areas.
2- Materials & Methods
2.1 Study area and dataset description
The present study was performed within the basin of the Caspian Sea.Drainage basin is considered to be a standard unit ofstudy in environmental studies and thus due to the applied nature of the present study, the Caspian Basin was selected as our study site. In this study, a new FCD model was implemented for data collected from Landsat 5(1366) and Landsat 8 (1396).
2.2 Proposed approach
In the present study, an improved FCD model was obtained by adding two steps to the initial FCD model. In the following paragraph, these two steps will be explained.
2.2.1 Removing thermal index
The first limitation of the initial FCD model lies in the fact that implementing this model for data collected bysensors without thermal band is impossible, because advanced shadow index in the initial FCD model is calculated by combining shadow and thermal indices. Thermal index is only used to separate the shadow of vegetation cover from black soil.In order to overcome this limitationin the improved FCD model, thermalindex is replaced with NLI and GNDVI indices. In this way, black soil and vegetation shadows are separatedusingacombinationofshadow, BI, NLI and GNDVI indices.
The NLI index can be calculated using(1):
(1)
The GNDVI index is also calculatedusing(2):
(2)
2.2.2 Local thresholding
In the initial FCD model,black soil identification and shadow index improvement (advanced shadow index calculation) wereperformedusingthresholdingand based on the combination of shadow and thermal indices.In this model, a number is selected as the threshold of the heat index, and shadow index pixels with values less than this threshold are considered as black soil.Obviously, it is practically impossible to define a threshold and calculate advanced shadow index for large scale areas.
Localthresholding is a much more accurate method of thresholding, which is also used in the improved FCD model.In this method, image received from the study site was divided into 14 sections and a suitable threshold value was selected for BI, NLI and GNDVI indices in each section to calculate advanced shadow index.
Moreover, different thresholds were selected forforests in flat areasand elevated areas.In this regard, digital elevation model of the region was used to separate low-altitude and high-altitude areas.
3. Discussion& Conclusion
Results indicated that the proposed improved FCD model has provided a more accurate estimate of forest canopy density as compared to the initial FCD model.
According to the results, the overall accuracy and kappa coefficient of the initial FCD model were 86.24% and 68.43%, respectively.However, the improved FCD model had an overall accuracy of 96.98% and a kappa coefficient of 92.31% which confirms improved performance of the model.
Moreover, the statistical analysis of changes in the canopy densityindicated that the total area of Hyrcanian forests increased by about 161,963 hectares from 1366 to 1396. This includes an increase ofabout 79, 50 and 33 hectares in Mazandaran, Gilan and Golestan provinces, respectively.
کلیدواژهها [English]
- FCD model
- NLI
- GNDVI
- Local thresholding
- Hyrcanian forests