The use of multiscale remote sensing imagery to derive regional estimates of forest growth capacity using 3-PGS

A number of process models now exist that estimate carbon and water vapor exchange across a broad array of vegetation. Many of these models can be driven with information derived from satellite sensors. In particular, a large number use the normalized difference vegetation index to infer spatial and temporal shifts in the fraction of visible light intercepted (f phi (p,a.)) by vegetation. We utilized a simplified process model (Physiological Principles Predicting Growth from Satellites), initialized with Advanced Very High Resolution Radiometer normalized difference vegetation index-derived estimates of f phi (p,a.) to estimate at monthly time steps photosynthesis, respiration, and aboveground growth of forest vegetation within a 54,000 km(2) region in southwestern Oregon. We had data available from 755 permanent survey plots to provide an independent estimate of forest growth capacity. In addition. we took advantage of a satellite-derived classification of 14 major forest types to investigate the extent that generalizations might be made about their respective productive capacities. From weather stations and statewide soil surveys. we extrapolated and transformed these sources of data into those required to drive the model (solar radiation, temperature extremes. vapor pressure deficit, and precipitation) and initialize conditions (soil water holding capacity and soil fertility). Within the mountainous region we found considerable variation existed within each 1-km(2) pixel centered on each of the survey plots. Even by excluding comparisons where local variation was high, model predictions of forest growth compared poorly with those estimated from ground survey (r(2) = 0.4). This variation was only partly attributed to variation in canopy f phi (p,a). Local variation in climate and soils played an equal if not greater role. When the sample plots were stratified into 14 broad forest types. within which growth potential varied similarly (coefficient of variation for each of the 14 types averaged 6%), a good relation between predicted and measured forest growth capacity across all types resulted (r(2) = 0.82, P less than or equal to0.01, SE = 1.2 m(3) ha(-1) yr(-1)). The implications of these analyses suggest that: (1) models should be rigorously tested before applying across landscapes: (2) accuracy in locating plots and in extrapolating data limits spatial resolution; (3) soil surveys in mountainous regions are inaccurate and difficult to interpret, (4) mapped vegetation classifications provide a useful level of stratification; and (5) remotely sensed estimates of canopy nitrogen status and biomass increment and canopy nitrogen status are needed to improve and validate regional assessment of growth. Crown Copyright (C) 2001 Published by Elsevier Science Ireland Ltd. All rights reserved.