Stratification of regional sampling by model-predicted changes of carbon stocks in forested mineral soils

Abstract

Monitoring changes in soil C has recently received interest due to reporting under the Kyoto Protocol. Model-based approaches to estimate changes in soil C stocks exist, but they cannot fully replace repeated measurements. Measuring changes in soil C is laborious due to small expected changes and large spatial variation. Stratification of soil sampling allows the reduction of sample size without reducing precision. If there are no previous measurements, the stratification can be made with model-predictions of target variable. Our aim was to present a simulation-based stratification method, and to estimate how much stratification of inventory plots could improve the efficiency of the sampling. The effect of large uncertainties related to soil C change measurements and simulated predictions was targeted since they may considerably decrease the efficiency of stratification. According to our simulations, stratification can be useful with a feasible soil sample number if other uncertainties (simulated predictions and forecasted forest management) can be controlled. For example, the optimal (Neyman) allocation of plots to 4 strata with 10 soil samples from each plot (unpaired repeated sampling) reduced the standard error (SE) of the stratified mean by 9–34% from that of simple random sampling, depending on the assumptions of uncertainties. When the uncertainties of measurements and simulations were not accounted for in the division to strata, the decreases of SEs were 2–9 units less. Stratified sampling scheme that accounts for the uncertainties in measured material and in the correlates (simulated predictions) is recommended for the sampling design of soil C stock changes.