Figure 1 outlines the basic approach that will be followed in the proposed effort. In terms of the goals and objectives of the ADRO 2 program, our research project will focus on: (1) generation of unique information products (aboveground woody biomass and soil moisture) from Radarsat, ERS and JERS data generated from the Alaska SAR Facility; and (2) evaluating the use of these unique radar data in modeling patterns of net primary and ecosystem production (NPP/NEP) and soil respiration.

The modeling of NPP/NEP will be achieved by two approaches: (a) provision of information on seasonal soil moisture conditions and biomass distribution which will be used as inputs into the production efficiency model (PEM); and (b) as a basis for the upscaling of tower-based measurements of NEP. In terms of estimating total soil respiration, we will use the soil moisture products to provide appropriate adjustments to successional-based models of soil respiration that are derived from field observations.

The results from the models that use the radar-derived products as inputs will be evaluated in two ways. First, we will compare the outputs from these models to field based observations of NEP and total soil respiration from specific sites over specific time periods. And second, in the case of NPP/NEP, we will compare the results to those produced from the PEM model. In this way, we will be able to evaluate the performance of this model compared to actual observations as well as against an existing model that has been exercised in boreal regions.

The modeling approaches used in this study will require inputs provided from other satellite sources than the radar imagery. These include the AVHRR inputs required to exercise the PEM model (e.g., seasonal patterns of vegetation index, biomass, surface temperature, air temperature and soil moisture), as well as higher resolution data products produced from Landsat/ASTER (e.g., land/forest cover, burn severity, patterns of vegetation regrowth in burned areas, patterns of NDVI variations within burned and unburned stands, and patterns of surface temperature variation within burned and unburned areas).

The development and validation of the finer satellite data products will be based on field work conducted in the study sites since 1992 and ongoing studies being carried out during the summers of 2000 to 2002. The field measurements of net primary production are being provided via eddy tower covariance measurements collected during selected time periods in the summers of 1997 to 1999 and continuously during the entire year of 2000. The field measurements of soil respiration were collected in 1996 and 1997 and additional measurements will be collected during 2000.

Details of the above activities are summarized in the following sections.

3.1 Study Area and Previous/Ongoing Research

The proposed study will be based on field and remote sensing data collected at and over study sites within the Upper Tanana River Valley in interior Alaska. The study will build upon the results of a number of previous and ongoing research activities in the study region.

Coordinated field and satellite remote sensing studies have been conducted in this region since 1991. The remote sensing studies have used a variety of satellite systems (AVHRR, Landsat, SPOT, ERS-1/2, Radarsat, and JERS) to monitor the effects of fire. Field studies have focused on the estimating patterns of biomass burning during fire (Kasischke 2000), as well as the influence of fire on patterns of soil moisture, temperature, and forest recovery (Kasischke and French 1995). Measurements have been made of soil CO₂ emissions (O'Neill 2000; O'Neill et al. 1997; Richter et al. 2000), and net CO₂, water and energy fluxes using eddy covariance techniques (Chambers et al. 1999; Chambers and Chapin 1999).

The intensive study site that will be used in the proposed research program contains areas that burned in 1987, 1994, and 1999. We will specifically focus on areas within the 1987, 1994, and 1999 burns as well as adjacent, unburned stands where tower-based eddy correlation flux measurements of net ecosystem energy and CO₂ exchange were collected in the summers of 1997 to 1999 (Chambers and Chapin 1999; Chambers et al. 1999). These towers were deployed at several locations within the burned and unburned stands for periods of 2 to 6 weeks during 1997 to 1999. In addition, two towers are presently being installed to operate continuously in a black spruce stand that burned in 1999, as well as in an adjacent unburned black spruce stand. These eddy covariance measurements are being collected through 2000 through research sponsored by the National Science Foundation and the California Institute of Technology.

In the summer of 1996, total soil respiration measurements were initiated in burned and unburned black spruce forests under funding provided by NASA and continued through 1998 under a NASA Global Change Fellowship (O'Neill 2000). These measurements were collected using a dynamic closed gas exchange system (EGM-1 environmental gas monitor and SRC-1 soil respiration chamber manufactured by PP Systems). Measurements were collected in paired burned and unburned stands, using stands that burned in 1987, 1990, 1994, and 1996. A limited number of field measurements were collected during July of 1996 and at weekly intervals from May through August 1997. In addition, selected sites were visited in the summer of 1998 to collect deep soil respiration measurements. In addition to the field measurements, laboratory measurements of moss NPP as a function of temperature and moisture and laboratory incubation measurements of organic soil respiration measurements as a function of temperature and moisture were also carried out. The field measurements of total soil respiration were then combined with the laboratory measurements to develop a model of heterotrophic soil respiration (O'Neill 2000). This model allows estimation of the seasonal soil respiration in burned and unburned black spruce stands as a function of: (a) the age of the forest stand (years after burn); (b) soil temperature; and (c) soil moisture.

Using funding provided by NSF/USDAFS through the Bonanza Creek LTER (for travel and the salary for a student for the summer), we will collect additional chamber measurements of CO₂ emissions in same sites where the eddy covariance towers are based at several times during the 2000 summer growing season to provide additional data for validation of the soil respiration models.

We have archived well over 60 ERS-1/2, Radarsat and JERS-1 images from the study area from 1992 through 1999. Additional ERS and Radarsat imagery has already been requested through research being performed under a grant to one of the investigators for this proposal (Bourgeau-Chavez) through the Alaska SAR Facility/International Arctic Research Center (ASF/IARC). The focus of the ASF/IARC grant is to develop approaches to using ERS/Radarsat imagery to monitor fuel moisture conditions in the study regions, specifically to improve the spatial/temporal resolution of fuel moisture codes used by the fire management agencies in this region. Previous NASA-sponsored research has shown that variations in ERS radar backscatter in both burned and unburned black spruce stands were closely correlated with ground-fuel moisture codes produced by the fire management agencies (Bourgeau-Chavez et al. 2000). As part of the ASF/IARC research, soil moisture measurements will be collected in many of the same burned and unburned sites that will be used for this study. The data and results produced by the ASF/IARC grant will be directly applicable to the proposed study (see section 3.2 below).

The proposed research project will be closely coordinated with other ongoing satellite-based research/data develop projects being conducted in the study region, including:

1. Through previous NASA/EPA sponsored studies (including the University of Maryland's Global Land Cover Facility developed through NASA's ESIP program), an archive of 8 Landsat TM/ETM scenes collected from 1986 through 1999 have been obtained for the study region;

2. A study in cooperation with Compton Tucker of NASA GSFC is being initiated to study the inter- and intra-annual variations in the AVHRR-derived NDVI signatures over burn scars in the study region.

3. We are working closely with three-recently selected NASA projects that will use the remote sensing data and field data collected in their studies as well: (a) John Townshend's GOFC project that will validate automatic algorithms to detect and map deforestation/reforestation. The study area has been proposed as a boreal forest validation site; (b) Ruth DeFries' Pathfinder project that is focused on providing satellite-based data products (MODIS, Landsat and AVHRR) concerning land-cover change and land cover characteristics to the research community; and (c) Eric Kasischke's (the PI of this project) IDS proposal on boreal forest fire emissions, that will merge satellite (Landsat) data from the study region with field data to model fire emissions. This project will produce Landsat-based maps of pre-fire forest cover and post-fire fire severity for all the study sites in this proposed research activity.

4. We have requested ASTER imagery to be collected over our test sites. The ASTER science team has agreed to this request.

5. A collaborator on this experiment (Nancy French of the University of Michigan) is carrying out research for her NASA Global Change Fellowship on the effects of fire on albedo measurements in boreal forests. As part of this research, she is conducting field measurements of vegetation cover to determine how vegetation regrowth after fire influences albedo. This study is focusing on scaling issues, e.g., how does one scale measurements collected in the field to those derived from single-date, fine resolution imagery (Landsat) to those produced from coarse-resolution sensors on a high temporal frequency (e.g., MODIS/AVHRR). The approaches developed during this project address many of the same scaling issues faced in this study.

6. We have held discussions with the Director of the GLCF (J. Townshend) and he has agreed to incorporate our radar data set and data products into this facility, making it accessible to the broader user community.

We will focus our analyses on radar imagery collected between the four-year period of 1997 to 2000 that corresponds with field measurements. We will produce the following data products from the radar imagery:

1. Estimates of biomass re-growth in the 1990, 1987, and 1994 burns. We will prepare aboveground woody biomass maps within the burned areas using ERS or Radarsat data data collected in 2000. Biomass data from the field will be collected during the early summer of 2001 prior to initiation of plant growth. We will collect tree diameter and density information for areas with significant willow/aspen regrowth using point-quarter sampling techniques and harvest and weigh individual trees to determine the relationship between diameter and biomass.

2. We will produce biomass maps of the mature black spruce stands using JERs radar imagery from the study sites and biomass prediction equations developed by Harrell et al. (1995).

3. Seasonal soil moisture maps for the 1994 and 1999 burns sites. These maps will provide relative soil moisture levels (e.g., from a gradient of low to high) and will be calibrated based on field measurements collected in the 1994 study site. We will adjust the soil moisture maps for levels of biomass regrowth based on analysis of 1999/2000 Landsat 7 imagery. The estimated biomass levels for each year will be used in inputs for theoretical scattering models (Wang et al. 1994; 2000) to provide the level of soil moisture attenuation that results from plant regrowth. For the 1994 burn, we will assume that rates of regrowth are lower in the first several years after the burn and for the past two years.

4. Maps of high soil moisture levels after significant precipitation events. O'Neill (2000) demonstrated that heavy precipitation events significantly influence seasonal patterns of soil respiration. We will examine the SAR imagery collected at various times to determine the timing of the significant precipitation events, and the rates at which the soil dried after they occurred.

5. Relative seasonal soil moisture maps for unburned black spruce stands. Like the seasonal maps produced for the burned stands, these maps will contain information on relative patterns of seasonal soil moisture.

As illustrated in Figure 1, two different modeling activities will be carried out (PEM and PEM-FS) that will utilize data products derived from Radarsat, ERS and JERS SAR imagery, as well as from other satellite sources. In addition, data from finer resolution systems (such as Radarsat, ERS, and Landsat) will also be used to scale the field measurements up to the same scale as the outputs from the two PEM models.

Our first activity will be to exercise the baseline PEM model developed by the University of Maryland (Goetz et al. 1999; Goward et al. 1994; Prince and Goward 1995; Goetz and Prince 1998; Prince et al. 1998). This NPP model uses satellite imagery to measure the following variables: (1) canopy interception of PAR (FPAR); (2) incident photosynthetically active radiation (PAR); (3) air temperature and humidity; (4) vapor pressure deficit; and (5) surface soil moisture, among others.

The PEM model consists of two stages of modeling; in the first, biophysical variables are estimated for the study area using a series of models that are driven with satellite data; in the second, these variables are used to force a semi-mechanistic primary production model. The first stage includes models for each of the variables listed above. These models can be modified or replaced without altering the overall model structure. We intend to assess the validity and accuracy of each of these biophysical variable estimation models using field measurements in burned and unburned patches. The second stage of modeling is based on the production efficiency of energy absorbed by vegetation, that is, the amount of dry matter production per unit absorbed energy (e, gC MJ^-1), more generally referred to as light use efficiency. This approach is based on the temperature effect on the quantum yield of photosynthesis (Prince and Goward 1995; Goetz and Prince 1998). We plan to use the eddy correlation measurements of instantaneous carbon assimilation and incident PAR under unstressed conditions (wet periods with higher temperatures, early morning etc) to obtain field measurements of e. The PEM algorithms can be adjusted to account for the local conditions using these direct field measurements. Once parameterized, the model will be tested against the daily, integrated estimates of NPP provided by the flux towers with appropriate allowance for soil carbon efflux (heterotrophic respiration).

Initially we will use procedures developed for NOAA's TOMS and GOES satellites to measure incident PAR, and 1 km AHVRR (and MODIS data when available) data to estimate the remaining variables. We will first compile satellite and field data sets that are coincident in time and location with the eddy covariance measurements for the study region. In addition we will make new air and surface temperature measurements in various sites in the summer of 2000 in order to collect a sufficiently large sample of the data necessary to compare with estimates of surface and air temperatures using the TVX technique which is used in the PEM. Infrared thermometers will be used to measure surface (skin) radiative temperatures and screened thermometers for air temperature.

The PEM will be exercised over a two to three year period (1998 to 2000) in order to provide a preliminary assessment of how fire influences the seasonal patterns of NPP. The AVHRR inputs for the PEM will be generated through the GLCF, which has developed an AVHRR data processing module and an archive of global 1 km data. In addition to the AVHRR inputs, we will also exercise the PEM with standard data products produced by MODIS.

The next step in our modeling approach will be to investigate using higher spatial resolution data sets as inputs to the overall PEM model. Initially, we will substitute the radar-derived biomass and soil moisture data products (for the region that surrounds the 1987, 1994, and 1999 burn sites) into the model and compare the model outputs to both the standard PEM product and to field-based measurements. The next step will be to substitute other high-resolution data products into the PEM model, including the following:

A baseline forest/land cover map for the entire study region will be produced from a Landsat TM scene collected in 1992. A preliminary version of this map has already been produced for the area containing the 1994 burn (Michalek et al. 2000) and it will be updated for the present study. Specifically, we will create additional training sets for supervised classification using historical photographs as well as field observations. Undisturbed forests in this region have extremely slow growth rates, so it is possible to provide accurate training sets based on observations made in 2000 for the 1992 image. For the area burned in 1987, we have produced a preliminary pre-burn vegetation map based on a 1986 Landsat TM image. This map will be further refined through analysis of historical photographs and additional field surveys. Location of the burned areas in the study region will be based on maps provided by the Alaska Fire Service that have been verified through analysis of the burn scar signatures visible on Landsat imagery collected in 1995, 1988, and 1999 (e.g., one year after the 1987 and 1994 burns and 6 weeks after the 1999 burn). The Landsat image will also be used to identify other non-forested areas, e.g., agricultural areas, rivers and gravel filled floodplains, alpine tundra, bare soil/rock, and glaciers.

The fine scale information provided by this field / air photo / Landsat mapping activity will be used to provide the model with a more detailed characterization of the surface properties within the study region that affect the estimation of canopy light absorption - a key variable (when coupled with air temperature) driving potential photosynthesis.

A similar approach will be taken using fine scale thermal observations provided by Landsat and ASTER (and possibly MODIS). Thermal properties of the land surface and split-window estimates of atmospheric water vapor content will replace those used to estimate these same variables that, up to now, have been inferred within the PEM using only the AVHRR observations. The finer resolution land surface temperature (LST) and near-surface humidity drive key processes within the PEM that affect the inference of photosynthetic stress terms including air temperature (T_a) and vapor pressure deficit (D). Air temperature is a particularly important variable because it modifies NPP via three independent mechanisms: (i) the potential photosynthetic rate as expressed through absorbed radiation; (ii) the autotrophic respiration rate expressed as a exponential function of the departure of air temperature from the long-term climatological mean; (iii) the vapor pressure deficit term, a function of air temperature and surface humidity, expressed through reductions in potential photosynthesis as a "stress" term. Thus, the inference of more detailed spatial variability in air temperature permitted by the higher resolution imagery, for example over different portions of various aged fire scars, is expected to have a significant impact on the modeled NPP. Systematic variation in radiometric surface temperature estimated by the thermal IR AVHRR channels has been demonstrated at the global scale as a result of orbital drift in the AVHRR acquisition time, which had a significant effect on the interannual variability of NPP (Goetz et al. 2000).

We are calling this new approach PEM Fine Scale (PEM-FS). As with the baseline PEM modeling results, the PEM-FS modeling results will be compared to field observed NPP, as well as to the outputs from the baseline run.

The final approach to model NPP will be utilize upscaling approaches to develop a landscape-scale estimate of NPP (see, e.g., the right side of Figure 1). In this approach, we will analyze the tower-based measurements to determine how NPP varies as a function of parameters that can be monitored using satellite imagery (e.g., seasonal patterns of surface temperature/moisture, land cover, burn severity, etc.). We will then develop an approach to spatially/temporally extrapolate the tower-based measures to a spatial scale that matches the model outputs. These up-scaled spatial extrapolations will then be compared to the estimates from the PEM and PEM-FS models over the time periods when the field data were collected.

The modeling of soil respiration will be utilize a straight-forward extrapolation approach. Based on her field and laboratory measurements, O'Neill (2000) developed model that shows the patterns of soil respiration follow a distinct trajectory following disturbance by fire, with a dramatic increase in decomposition in the first few years following a fire following by a gradual decrease as the forest ecosystem recovers. O'Niell (2000) showed that patterns of soil respiration were influenced by: (a) rates of vegetation recovery at the site; (b) seasonal patterns of soil temperature; and (c) seasonal patterns of soil moisture.

Based on these observations, we propose to develop first-order model of the spatial/seasonal variations in soil respiration in the black spruce forests found in the study site. Based on the land/forest cover map, we will stratify the site into burned/unburned forests. To estimate woody biomass in the burned sites, we will use the radar data products. To estimate non-woody biomass in the burned sites, we will obtain Landat ETM data collected in the middle of the growing season for 2000 and produce NDVI maps for each of the burned sites. These NDVI maps will be correlated with the biomass maps collected for the radar studies to relate NDVI to biomass (which should be possible for the low biomass levels found in the 1987, 1994 and 1999 burned sites). For this study, we will assume that biomass recovery in the burned sites follows a linear trajectory as a function of time (in years) since disturbance. We will then use the variations in biomass within each burn to adjust the seasonal respiration estimate accordingly.

Next, we will make adjustments to the seasonal respiration estimates according to soil moisture conditions. We will use the radar derived soil moisture estimates to identify relative levels of soil moisture within the burned and unburned stands and adjust the seasonal respiration estimates according to the relationships identified by O'Neill (2000), who found respiration to be positively correlated with soil moisture.

Finally, we will adjust the soil respiration measurements based on spatial/ temporal variations in surface temperature observed from the different satellite systems. O'Neill (2000) showed that soil respiration is positively correlated with soil temperature. O'Neill (2000) developed seasonal profiles of soil temperature to the seasonal permafrost profile (in mature black spruce stands) and to depths of 1 m in burned stands. The assumption we will use in our modeling effort is that the seasonal soil moisture profiles will be linearly proportional to seasonal variations in surface temperature. We will develop an appropriate scaling factor to apply to the soil temperature profiles developed by O'Neill (2000) based on satellite observations. We will then adjust soil respiration accordingly.

As with the other parts of this modeling effort, we will compare the model predictions to field observations for the appropriate time periods.