SigRes HoughtonR Jan2004

Slide1: 

Carbon in the Forest Biomass of Russia R.A. Houghton Olga Krankina Tom Stone Warren Cohen Peter Schlesinger Thomas K. Maiersperger David Butman Doug Oetter Woods Hole Research Center Oregon State University

Slide2: 

Context

Context: 

Context 1. What is the C balance of the northern mid-latitudes? 2. What are the mechanisms responsible for the current (and future) C balance? a. Are forests growing faster? (physiology) b. Are more forests in a regrowth phase? (age structure – past disturbances - LCLUC)

Slide4: 

How much carbon is in the biomass of Russian forests? How has that amount changed in the last decade(s)?w Questions:

1990 Forest Cover Map of USSR: 

1990 Forest Cover Map of USSR What would the map look like in 2000?

Slide6: 

Approach

Approach: 

Approach Russian forest inventory data for training Landsat ETM+ (growing stock -> C/ha) Landsat ETM+ for training MODIS MODIS for scaling to all Russia

Stratify Russian forests into ~15 ecoregions by…: 

Stratify Russian forests into ~15 ecoregions by… Geo-regions (4): European Russia, Western Siberia, Eastern Siberia, Far East Vegetation zones (5): Northern, central, southern taiga, temperate forest, forest steppe

Slide11: 

Forest Inventory Data

Slide13: 

Total Biomass predicted from growing stocks …by age …by eco-region …by species (group)

Slide14: 

Training Landsat data with Inventory Data

Slide15: 

Videl (inventory polygon) data overlaid with Landsat ETM+ data

Fitted function: 

Fitted function In theory… i

Slide17: 

Observed & predicted biomass for individual videls (polygons) In fact…

Slide18: 

Inventories acquired, processed, …

Another test (coarser resolution)…. : 

Another test (coarser resolution)…. Compare larger forest inventory unit (lesnichestvo) with Landsat-derived estimates of ... Forest area Average C/ha

Slide20: 

Scaling up with MODIS

Approach: 

Approach Russian forest inventories for training Landsat ETM+ Landsat ETM+ for training MODIS MODIS for scaling to entire Federation

Slide22: 

But wait…

An Alternative Approach: 

An Alternative Approach At a more aggregated (coarser) scale (leskhoz), MODIS may be a reasonable predictor of biomass C (??)

Slide24: 

Leskhoz boundaries ~1880 Leskhozes in the Russian Federation

Slide25: 

MODIS surface reflectance (MOD43B4) (BRDF product)

Slide26: 

Leskhoz boundaries ~1880 Leskhozes in the Russian Federation

Slide28: 

Thank you

1990 Forest Cover Map of USSR: 

1990 Forest Cover Map of USSR

Geo-Ecoregion stratification for sampling: 

Highly generalized vegetation zones of the FSU based on the work of Kurnaev. Forested zones - green. The mixed forest (light green). The brown zone - forest steppe, which grades south into steppe. The four regions of Russia to be studied are colored and are (l to r), European-Urals, West Siberia, Central Siberia, and Eastern Russia. Geo-Ecoregion stratification for sampling

Slide35: 

GLC 2000 Forest Cover for Russia (8.3 million Km2) MODIS IGBP Forest Cover for Russia (6.5 million Km2)

Slide36: 

Spatial Analysis for the Development of a Russian Biomass Map Iterative Iso-clustering of Digital Data: ERDAS: Unsupervised Classification of Atmospherically corrected DN’s into forest and non-forest land cover classes Non-forest includes all agriculture, urban etc. Recoding the raw DN’s into a forest sub-type classification: -Pine -Spruce -Mixed Conifer -Deciduous -Mixed Forest Training plot derived from this classified data coded 1-5 for each forest type The recoding of the digital data is based off of the percent composition data given within the Forest Inventory. Supervised Classification: ( Maximum likelihood Classification) Derive classified images that are specific to the forest types: Pine Spruce Mixed Conifer Deciduous Mixed Forest Have in hand separate spatial datasets. These datasets are used to separate the image into batches of data that reflect each forest type.

Slide37: 

Rule: (Due to Co-linearity between Bands) Use two bands from the visible spectrum Bands 1 or 2, and 3 Use two bands from the infrared spectrum Bands 4, and 5 or 7 Bands Selection and Statistical Analysis Done By Species For the Development of A Russian Biomass Map Split Forest Inventory Data set into a Testing and a Training Dataset: Model 1 Model 2 Analysis of Raw DN’s Perform Transformation on all bands Square Root Square Log Inverse Log To determine if non-linear relationships exist and can be corrected for using various techniques. Create Correlation Plots for all bands and all transformations against Vegetation data: (Biomass). The correlation plots between raw DNs and biomass are used to determine if non-linear relationships exist, and then the transformations are created and re-plotted to see if they correct for the non-linearity. R-Square: Develop a ranking mechanism to tell which combination of bands/transformations produces the strongest correlations with the Biomass Values Canonical Correlation Analysis: Evaluate the four bands and the inventory biomass values using a CCA technique. Derive Canonical Index of landsat values: Reduce Major Axis Regression: (RMA) Used to perform a regression between the inventory biomass values and the summed Canonical index values From the RMA we obtain an equation that is then multiplied against the original CCA Biomass Indices to create a biomass layer that is species specific. CCA INDEX Derivation: The index is created by summing the Input bands/transformations, that have been Standardized about the mean and multiplied By the canonical coefficient for that band RMA Technique: Slope = (sign of correlation +/-) Sdy / SDx x = LANDSAT (CCA Index) y = BIOMASS (Inventory) Intercept = mean Y - slope * meanX Pred = slope (LANDSAT) + intercept RMSE = square root ( average (residual squared) ) )