Landuse change is an important aspect of global carbon analysis. Extensive efforts have been devoted to modeling the effects of landuse change on global carbon cycle, however, most of efforts have focused on disturbance effects. The direction and rate of forest succession processes have received far less attention, although they are critical components of modeling carbon flux over longer time periods. How disturbed ecosystem recover has become a global question, and yet little is known about the rates and controls of secondary forest succession. To begin addressing this problem, I propose to study secondary forest succession process with the Pacific Northwest (PNW) as the research area. Characterizing the secondary forest succession after disturbance is important because both live biomass and detritus accumulation rates are highly dependent on the life forms present (i.e., herb, versus shrubs versus hardwoods versus conifers). Current understanding about forest succession and its role in controlling global carbon sinks and sources could be improved by (1) additional analysis of the rate, spatial and temporal pattern of secondary forest succession. (2) development of spatial model based on empirical observations from remote sensing. The proposed research project will address both aspects by linking remotely sensed and ground-based biogeoclimatic data to early forest succession model which will provide detailed information for carbon budget analysis.

Pacific Northwest is a great place to develop the role of secondary forest succession in carbon cycle analysis, and has high potential to store C per unit area, several times higher than the global average (Harmon et al. 1990). In the classic view, early succession following wildfire in the Oregon Cascades region generally proceeds through stages of herbs and grasses, then broadleaf shrubs and seedlings, and finally reach young canopy-closed Douglas fir (Franklin and Dyrness, 1988). Recent change detection research, however, indicates that succession trajectories differ significantly over extensive areas, which may lead to vastly different rate of carbon accumulation (Cohen et al, 1996). Mature forests in the Pacific Northwest contains great amount of stored carbon, and the forest has experienced and will experience substantially changes due to human disturbance related to logging (Cohen et al, 1998), which releases much of the stored carbon into atmosphere. The study area of the proposed project ranges from the west side of the Cascade across the Willamette Basin to the west side of the coast range, which is covered by the Landsat TM scene 46/29.

The main objectives of this project are to answer three questions concerning forest succession after stand replacement: (1) How similar are secondary succession trajectories in Western Oregon in terms of rate and direction? (2) Which spatial and environmental factors are associated with these successional trajectories? (3) What are the spectral changes associated with the various succession trajectories?

To answer these questions, I will: (1) develop contemporary and historic forest succession data sets, from which early secondary forest succession variability in Western Oregon can be identified and characterized; (2) develop an empirical succession model which synthesize the succession trajectories and potential controlling factors; and (3) develop spectral changes associated with different succession trajectories and map the historical succession trajectories with Landsat TM and MSS data. The latter allow me to validate the empirical model, and to predict future succession trajectories.

Remote sensing is a promising tool for direct observation of forest succession. The quantitative assessment of the forest succession processes from remote sensing can be combined with a variety of potential controlling factors of secondary forest succession in an empirical diagnostic model. The proposed research will answer the "when" (logistic analysis), "why" (regression tree analysis), and "where" (spatial statistics) of forest succession trajectories.

The proposed research will contribute to a project funded by NASA Earth Science Enterprise LULCC (http://www.fsl.orst.edu/lter/research/hjarel/russia.htm), whose overall objective is to determine the relative importance of land-use versus biogeoclimatic factors in controlling spatial and temporal patterns of carbon dynamics. This proposed research is closely related to the LandCarb model (Wallin et al, 1996). First, it will develop new and improved forest succession data sets in western Oregon from historical aerial photographs and satellite images, which will provide spatial extent and location of succession trajectories for the LandCarb model. Second, by using spatial and temporal explicit data sets, and spatial biogeoclimatic data, a spatially explicit, empirical succession model will be developed, which will provide important new information about how forest succession influences carbon storage in Pacific Northwest. Finally, it will provide a general framework for study of secondary forest succession using remotely sensed data.

The proposed research will be organized around two working hypothesis: (1) Early secondary forest succession is controlled by the physical environment (geo-climatic), stand initialization, and biological interactions. (2) Spectral changes associated with different early forest succession trajectories follows distinct pattern of progression through spectral feature space.

The identification and characterization of early forest succession trajectories are needed to link with biogeochemical models to predict their influence on the carbon budget. Direct observation of forest succession with satellite remote sensing and aerial photography identify areas prone to prolonged shrub and hardwood occupancy versus those that are rapidly occupied by conifers. Important data sets to be acquired for this analysis include: (1) 30 meter resolution DEM data set, (2) contemporary vegetation, (3) aerial photography, (4) stand origin data. The required data sets will be acquired from a number agencies. The DEM data set will be obtained from USDA Forest Service at Corvallis. The contemporary vegetation map was developed by Dr. Warren Cohen from TM imagery. The land ownership of the study area belongs to a number of agencies including Bureau of Land Management, USDA forest Service, and private owners. Aerial photography and site origin will be obtained from these agencies. Initial investigation indicates aerial photographs of most of the study area date back to the 1940’s.

Based on a contemporary vegetation map, the nonforested area will be masked out on the DEM data sets. Preliminary analysis in H.J. Andrews Experimental Forest in the Oregon Cascades indicate that topographic aspect and elevation are important factors in empirical succession model (Nesje & Cohen, 1996). I will therefore classify the study area into 24 classes based on aspect (8 subclasses) and elevation (3 subclasses: <900m, 900m-1200m, >1200m) from the DEM. Stratified random samples (7200 samples totally) will be selected from these classes. Once located on aerial photographs, photo interpretation will be used to determine information about the percentage cover of the different lifeforms (herb, versus shrubs versus hardwoods versus conifers) for various times since disturbance. Combined with stand origin data, information about time to reach 75% conifer cover and cumulative conifer cover over time will be calculated. Statistically distinct succession trajectories will then be identified by overlaying harvest date and the above information, and tracking succession development from harvest to closed canopy conifer condition.

Based on the working hypothesis 1, development of an empirical model for observed forest succession trajectories in association with the potential controlling factors is the focus of this research activity. The observed succession trajectories serves as input to Markov chain models or/and logistic function models, which will simulate the rate forest succession from one lifeform to another. Regression tree analysis (Efron & Tibshirani, 1991) will be applied to relate the succession trajectories to the hypothesized driving forces. Spatial statistical analysis, which incorporates remote sensing, GIS, and a multivariable, multi-temporal mathematical model, will form the third part of the empirical diagnostic model. This will allow one to analyze forest succession trajectories in relation to geographically referenced data to determine the spatial pattern of the succession process. The final empirical succession model will be then applied to the whole study area to develop a map for the succession trajectories in a prognostic way.

It is the focus of this part to develop the spectral trajectories, which then can be used to extract information about the actual succession patterns to validate the final model of focus 2. The spectral trajectories can also be used to predict future succession pattern. I hypothesize changes in forest succession follows a pattern of progression through spectral feature space. By adding a temporal dimension into the spectral feature space, spectral trajectories associated with different succession trajectories will then have distinct patterns.

Data needed for this analysis are mainly the multidate satellite imageries, the USGS DEM, and aerial photographs. Multispectral Scanner scenes of 1972, 1977, 1984 and Thematic Mapper scenes of 1988, 1991, 1995 of the study area are available from my advisor Dr. Warren Cohen. USGS DEM and aerial photographs are available from focus 1. All the images have been geometric rectified with each other at Dr. Warren Cohen’s lab. Scanned aerial photographs of the selected forest stands within each of the identified succession trajectories will be rectified with the satellite images. The images will be radiometrically rectified to convert the digital number values to absolute radiance values based on transformations derived from scene invariant elements of either very low or very high albedo (Hall et al., 1991). Tasseled Cap transformation (Crist and Cicone, 1984) will be calculated from the pre-processed data, and tasseled cap of the selected stands will be extracted using Universal Transverse Mercator (UTM) coordinates, averaging the values for a 3 pixel by 3 pixel block surrounding each site. The spectral properties of the distinct succession trajectories will be analyzed using ordination in spectral space. Statistical analysis will be performed to test the validity of the spectral trajectories. At the same time, spectral mixture analysis will be performed with endmembers (Adams and Smith, 1986) derived following Bateson A. et al (1996). Spectral mixture analysis will then be performed on the different succession trajectories. The spectral information obtained above will then be applied to the whole study area, and hence information about the actual early secondary forest succession trajectories can be determined, which will be compared with predictions from the model developed in focus 2. By fitting the observed spectral change to the spectral trajectories of the different secondary succession trajectories, the future succession trends and rate can be projected.

First, the proposed research will provide better understanding of the controls of secondary forest succession in terms of rate and direction. This new knowledge about forest recovery process after disturbance will be important information to the global change research community. Second, the developed succession model when run in prognostic mode will directly contribute to the LandCarb model by providing information about spatial extent and location of these successional trajectories. Third, the research may also form a general framework and methodology for future forest succession studies using remotely sensed and spatial biogeoclimatic data. Finally, data about forest succession process would be valuable resources for wildlife habitat analysis and ecosystem management.

Wallin, David O.; Harmon, Mark E.; Cohen, Warren B. [and others]. 1996. Use of remote sensing to model land

Global carbon cycle is one of the most important questions faced by ecologist, and Landuse change is an important aspect of global carbon analysis. Most efforts on forest cover change have focused on disturbance effects, while little is known about the rates and controls of secondary forest succession. How disturbed ecosystem recover has become a global question due to human activities. The proposed research will address "when", "why", and "where" early secondary forest succession trajectories proceed in western Oregon by linking remotely sensed, ground-based biogeoclimatic data, and human factors. There are three major tasks in the proposed research: (1) develop contemporary and historic forest succession data sets to quantify succession trajectories. (2) develop an empirical succession model which can be used to predict the succession process after disturbance, and (3) identify spectral trajectories associated with different succession trajectories, which can be used to validate the empirical succession model, and predict future succession trend.

The proposed research will contribute to a project funded by NASA Earth Science Enterprise LULCC (http://www.fsl.orst.edu/lter/research/hjarel/russia.htm), whose overall objective is to determine the relative importance of land-use versus biogeoclimatic factors in controlling spatial and temporal patterns of carbon dynamics. This proposed research is closely related to the LandCarb model (Wallin et al, 1996). First, it will develop new and improved forest succession data sets in western Oregon from historical aerial photographs and satellite images, which will provide spatial extent and location of succession trajectories for the LandCarb model. Second, by using spatial and temporal explicit data sets, and spatial biogeoclimatic data, a spatially explicit, empirical succession model will be developed, which will provide important new information about how forest succession influences carbon storage in Pacific Northwest. Finally, it will provide a general framework for study of secondary forest succession using remotely sensed data.