Forest Canopy Density Mapping of Doon Valley using Biophysical Response Modeling on MODIS data
Dr. Sharad Tiwari
Tropical Forest Research Institute
Mandla Road, Jabalpur (M.P.), 482021
2Indian Institute of Remote Sensing
Department of Space
Forest Canopy density is a major factor in the evaluation of forest status and is an important indicator of possible management interventions. FCD mapper is one of the useful tools to detect and estimate the canopy density over large area in a time and cost effective manner. This model is based on four indices i.e. soil, shadow, thermal and vegetation. This model requires very less ground truths, just for accuracy check. In the present work, we have tested the FCD model in Doon valley. Forest canopy density map of Doon valley has been generated using MODIS and ETM+ images. The overall accuracy for ETM+ was 72% and for MODIS was 61%. Comparison of the forest area mapped by ETM+ and MODIS shows that with MODIS data, the overall forest area mapped is over estimated as compared to forest area mapped with ETM+. This is due to coarser resolution of MODIS. Because of coarser resolution, small gaps inside forest vegetation are not being detected and are mapped into forest. Therefore, forest canopy density mapping using bio physical response modeling on MODIS data will help to generate a map of forest density status having more number of density classes with a compromising accuracy in a time and cost effective manner.
Forest Canopy density is a major factor in the evaluation of forest status, as it is directly or indirectly related to the problems arising due to various factors like soil erosion, biodiversity loss and climate change. Many studies performed earlier, show that canopy gaps widen due to fall of trees or branches. Some other activities like exploitation of natural resource, urbanization, increasing population, grazing etc also affects, which in turn affect the local ecosystem, environment and global biosphere. Therefore, study of forest canopy density forms a launching pad for the policy makers in formulating developmental plans. This will provide great help in the prioritization process and planning management strategies of forest on sustainable basis.
Study Area and source data
The study was conducted at Doon valley. The area lies between 30° 2´ 31˜ to 30° 26´ N latitude and 77° 52´ to 78° 19´ E longitude covering a forest area of 555.5 km2. Site is slight hilly as well plain. Forest cover occurs at the middle and foothills are mostly dominated by Sal (Shorea robusta) forests. The other species occurring in this part are Saint (Terminally balata), Bali (Anogeissus latifolia), Sain (Terminalia tomentosa), Haldu (Adina cordifolia), Bahera (Terminalia balerica), Jhingan (Lannea coromandelica), Semal (Bombax ceiba) etc. The source data are MODIS data of Doon valley of 4th November 2003 and Landsat Enhanced Thematic Mapper (ETM+) data of 14th Oct, 2002.
In the present study, remote sensing and geographical information system as well as extensive fieldwork has been carried out for density mapping. The outline of the analysis is shown in Fig. 1. For ground truth collection, forest canopy density was measured using ocular method and locations were recorded using GPS. The overall area was classified into four density classes i.e. 10% to 30%, 30% to 50%, 50% to 70% and more than 70%. This way, data were recorded in total of 27 plots. At each plot five observations were recorded.
Fig 1: Logical Flow Diagram of work
ETM+ image for the year 2002, Oct. 14 and MODIS image for the year 2003, Nov. 4 has been used for the study. ETM+ image has been georeferenced using toposheet. Then MODIS image has been georeferenced using image to image registration method, taking geocorrected ETM+ image as master image. Since FCD mapper supports ETM+ data i.e. 7 bands data, where as MODIS contains 36 bands, therefore the equivalent bands of MODIS to that of ETM+ image were chosen and then these 7 bands selected out of 36 bands, were stacked (Table 1). All seven bands of MODIS image were converted to BSQ format for further processing with FCD mapper. Since MODIS data is of 12 bit unsigned integer type, the stacked image is rescaled in to 8 bit unsigned integer as FCD supports the 8 bit unsigned integer data type only.
Table 1: Band combination for MODIS and Enhanced TM+
Forest Canopy Density map was generated using FCD mapper based on FCD model (Rikimaru, 2000). The stacked image of MODIS of these seven bands was used for the FCD mapping and all the seven bands were taken. Values were assigned for the record length and for the end pixel. Finally, images were normalized and AVI was chosen in PCA option. Similarly TI (Thermal Index) was normalized. Furthermore, normal equation was chosen for generation of SI (Shadow Index) and gap detection was chosen to obtain ASI (Advanced Shadow Index). The threshold values for different parameters were assigned for water set, points for gap set. Similarly, the threshold points for the vegetation set, Bare Soil set, minimum vegetation density and maximum vegetation density were set. After finalizing the setting of the threshold values, the software automatically starts to calculate VD. At the same time SSI (Scaled Shadow Index) was generated from VI-BI-SI indices. Finally forest canopy density model was completed by the integration of VD and SSI. This integration process is shown in Fig 2.
Fig. 2: Procedure of FCD Mapping Model
Result and Discussion:
In the present work, FCD model has been applied on ETM+ as well as MODIS data, for generating forest canopy density maps. The accuracy of the work has been evaluated using 135 numbers of sample plots collected from 27 test sites. These points are distributed in different density classes for the study area. The canopy density has been analyzed based on the classified map generated form FCD mapper (Fig 3 for ETM+ and Fig 4 for MODIS). In FCD mapper, the map has been generated in eleven classes at an interval of 10%. However, in the field we have collected data in four density classes i.e. 10-30, 30-50, 50-70 and more than 70%. Then the accuracy assessment was performed. For this purpose, confusion matrix was created and overall accuracy and kappa statistics are calculated. The over all accuracy for ETM+ data has been 72% and for MODIS data it has been 61%.
Figure 3: Forest Canopy Density Map of Doon Valley (ETM+ Data)
Figure 4: Forest Canopy Density Map of Doon Valley (MODIS Data)
On comparison of the forest area mapped by ETM+ and MODIS (Fig.5), it is found that overall forest area generated with MODIS data is higher then that of ETM+ data. This is because of coarser resolution of MODIS. Due to coarser resolution and comparatively smaller study area, MODIS is unable to detect small gaps inside the forest vegetation.
Fig 5: Forest Area as differentiated by ETM+ and MODIS data
In the present study, ETM+ and MODIS data have been used for calculating the forest canopy density based on biophysical response model. For ETM+ data the accuracy achieved was 71%, which is quite acceptable. However, the accuracy with MODIS data achieved was 61%. The less accuracy obtained with MODIS may be due to the comparatively small area for MODIS, as it has coarser resolution, ranging from 250 meter to 1 KM. So our recommendation is that, for MODIS image, the study area should be larger. Result reveals that, the approach is very good and can be very useful for the agencies, which are involved in the canopy stratification work, to look at the global and regional level status of the forest area.
- FAO, 2001, State of the Worlds Forests, www.fao.org/ docrep/ 003/ y0900e.
- Roy, P. S., 1999, Space remote sensing for forest management, FCD user manual guide, Japan, ITTO/ JOFC Association, Japan, 2000.
- Panta, Menaka, 2003, Analysis of forest canopy density and factors affecting it using RS and GIS techniques, M. Sc. Thesis, ITC Netherlands.
- Roy P. S., Miyatake, S. and Rikimaru, A., 1997, Biophysical Spectral Response Modeling Approach for Forest Density Stratification, www.gisdevelopment.net/ aars/acrs/ts5/ts5008pf.htm.
- Rikimaru, A. and Miyatake, S., 1997, 18th ACRS, Kuala Lumpur, Malaysia, Oct-1997.
- Rikimaru, A, 1996, Utilization of RS in site assessment planning for rehabilitation of logged over forests. Paper presented at the proceedings of workshop, Thailand, Bangkok.
- Zhu Zhiliang, 1994, Forest Density Mapping in the Lower 48 States: A Regression Procedure, www.srs. fs.usda.gov/pubs/.
- Jensen, John R., 1996, Introductory digital Image Processing, Prentice Hall, NJ.
- FCD Mapper, 1999, User Guide V 1.0, ITTO, Japan Overseas Forestry Consultants Association, Jun 1999.
- Parker, Geoffery and Brown, Martin J., 2000, Forest canopy Stratification, The American naturalist, Vol. 4, No. 155, pp 473-484, April 2000.