Insights in the Forests Role in the Carbon Cycle and Management Criteria. Differences between a Protected Area and a Sustainable Manged Forest. RESULTS

In this work, we assessed the past, present and future health of vegetation (as quantified through the NDVI) in two different areas with different management plans. One of them (Fragas do Eume) consist of a protected area, regarded as an example of the climax vegetation in an Atlantic Rainforest. This area mostly consists of forests of native deciduous species (A. glutinosa, Q. robur, C. sativa or B. alba) with limited resource usage. In the other hand, the other area (Baroña) is governed under the UN definition of SFM, allowing the extraction of resources and promoting a sustainable usage of them. The main economic activities are resin extraction, grazing, silviculture and apiculture, which can shape the vegetation cover and type of vegetation. Thus, Fragas do Eume shows a higher value of NDVI (0.7410), more constant in time (Kendall τ = 0.0376) with a marked seasonality and highly correlated with environmental variables. In comparison, Baroña shows a lower average value of NDVI (0.6839), with an apparent positive trend (Kendall τ = 0.2933), due to the recent increase in the social awareness on the sustainable use of the area. Furthermore, NDVI shows a weaker seasonality in Baroña, with a double peak at ~0.5 and ~1-year periods. This is due to the different type of vegetation cover (forests of P. pinaster, P. radiata, E. globulus; and grazing areas) derived from the economic uses. This defence of  the vegetation health in Baroña with anthropogenic activities causes a lower correlation with environmental variables, making it more challenging to develop a predictive model to forecast the NDVI in the future. Thus, we could predict the NDVI in Fragas do Eume with an RMSE of 6% of the NDVI range (R² > 0.9) in the training stage and 9% (R² ~ 0.9) in the testing. Nonetheless, in Baroña we obtained errors from 6 to 14% (R² = 0.7 - 0.9) at training, but RMSE of 15 – 23% (R² ~ 0.5) in the testing. Generally, the most accurate algorithm was the Random Forest, but Bayesian Neural Networks performed similarly and faster.

Sites description

• fde: Fragas do Eume. A protected area (National Park) in the province of A Coruña (Galicia Spain), representativo of the climax vegetation in an Atlantic thermofile rainforest.
• barona: CMVMC of Baroña, an area collectively owned and managed by the inhabitants in accordance with the UN principles of Sustainable Forest Management (SFM).

Scenarios description

• RCP26: Representative concentration pathway with 2.6 W m^-2 of radiative forcing for the year 2100. It represents the "very strict" emissions scenario in which the CO2 emmissions start declining by 2020 and reach zero by 2100.
• RCP85: Representative concentration pathway with 8.5 W m^-2 of radiative forcing for the year 2100. It represents the "business as usual" scenario in which the emissions contunue rising during the whole 21st century.

Files description

• 1-MK-test_<site>: Results of the modified Mann-Kendall test for the trend.
• 2-WT-analysis_<site>: Results of the wavelet transform analysis for the seasonality.
• 3-CCF_<site>_<scenario>:  Results of the cross-correlation analysis between NDVI and environmental variables.
• 4-Training_<site>_<scenario>: Performance of the different algorithms in the training stage for NDVI prediction.
• 5-Testing_<site>_<scenario>: Performance of the different algorithms in the testing stagefor NDVI prediction.
• 6-Forecast_<site>_<scenario>: Forecasted NDVI series.