The primary objective of our study was to develop a spatial model to quantify forest productivity across a mixed, temperate forest using a suite of remote sensing, ancillary environmental and yearly climate data. Prior research has demonstrated that satellite-based, landscape scale assessments of forest productivity can be improved with the addition of climate data. However, these studies were primarily focused on homogenous forests, with limited field data to assess accuracy across diverse landscapes. To better understand how accurately we can predict forest productivity across a large, complex forested region, and the sensitivity of these models to predictor variables, we compared productivity models calibrated with 10 years of basal area increment measurements at 132 field sites across Vermont and New Hampshire, USA, representing seven different tree species. Model input parameter configurations included:
- MODIS seasonal NDVI and NDWI metrics only,
- MODIS metrics, plus ancillary site, stand and habitat quality spatial data layers,
- MODIS, plus ancillary, plus PRISM-based monthly climate metrics
Results/Conclusions
Results indicate that overall, it is difficult to model productivity across a heterogeneous landscape. This is likely due to differences in spectral signals based species, stand density and understory composition that is independent of yearly variability in productivity. However, satellite-based models did improve significantly with the addition of ancillary environmental (r2 from 0.01 to 0.31 and AIC from 9547 to 8333). The addition of climate predictor variables only marginally improved the model further (r2 0.40, r2Adj = 0.37, AIC 8325). This indicates that in northeastern forests, the addition of site conditions can significantly improve satellite assessments of forest productivity, by primarily quantifying environmental stressors such as acid deposition and site quality. The limited improvement with climate metrics is likely because climate is not a primary driver of productivity in this relatively humid, temperate environment where species are well adapted to climate conditions.
The results of these analyses inform best practices for modeling forest productivity but also highlight some of the potential drivers of spatial and temporal variability in forest productivity in northeaster forests. This information is of great interest considering the role of temperate forests in carbon sequestration and desire to understand spatial and temporal patterns across the landscape.