A paper co-authored by Space Intelligence’s Chief Scientific Officer, Dr Edward Mitchard has shown how large scale forest mapping products can underestimate the level of deforestation occurring.
The team, led by Dr David Milodowski of the University of Edinburgh’s School of GeoSciences, compared three forest loss products over Brazil: Global Forest Watch (30m resolution); PRODES (60 m resolution) and FORMA. They compared these products with the results of their own analysis. This involved taking the multi-year differences in classified RapidEye satellite images. The RapidEye satellite has a spatial resolution of 5m, and as such was used as the ‘truth’ dataset.
The team found that broadly the general spatial patterns of change were consistent across the products. However, the marked differences arose when taking into account the size of the forest disturbance being mapped. Smaller clearances of less than 2 hectares were less well mapped by the wall-to-wall products from GFW and PRODES. This suggests that large scale mapping using these products, especially in areas where forest change is dominated by small scale clearances, will result in significant negative bias, or underestimation of deforestation rates.
In order to account for these differences in ability to measure small scale deforestation, the authors suggest that analysts apply a correction factor of +25% for the GFW product, and +50% for the PRODES product, in order that the results can be more directly comparable with those achieved through higher resolution mapping at 5m. In areas where deforestation is dominated by large scale clearances, the authors suggest that a smaller correction factor of 5% for GFW and 15% for PRODES should be applied. Yet they counsel caution in the uncritical application of such correction factors, since these results are preliminary.
Dr Milodowski said:
“Our study shows that the performance of satellite-based deforestation detection depends on the style of deforestation occurring in a given area: small disturbances and degradation are much harder to detect than large clearings. In turn, these reflect the underlying drivers of deforestation. This also makes it challenging to generalise correction factors across areas with different forest characteristics and different drivers of deforestation, but in regions where small scale clearance and degradation dominate underestimates are likely to be particularly significant.”
The authors conclude by noting that that the differences between products are not errors as such, rather they reflect differences in forest definitions and aims of different monitoring systems. At Space Intelligence we are able to advise on the best solution for your monitoring requirements, and have provided some more information on the differences between our own forest monitoring product and the GFW product here: Why Use Space Intelligence to Monitor Forests?
Reference
Milodowski, D.T., Mitchard, E.T.A. & Williams, M. 2017. Forest loss maps from regional satellite monitoring systematically underestimate deforestation in two rapidly changing parts of the AmazonEnvironmental Research Letters, https://doi.org/10.1088/1748-9326/aa7e1e
