Browsing by Subject "matching"

Measuring the effectiveness of protected areas (PAs) is essential as they are key tools in tackling the ongoing biodiversity loss and there is substantial variation in their effectiveness (the estimated ability of protected areas to prevent unnatural disturbances). In forested PAs, the most common variable in effectiveness estimation is forest loss, but fire can also be used as a proxy for conversion. There is, however, a lack of robust comparisons between different data sets and proxies. This thesis aims to provide more insight into the issue by comparing three satellite-based data sets in protected area effectiveness assessment using Madagascar as a case study. The questions to be answered here are whether the data sets and variables derived from them produce similar PA effectiveness estimates and whether they could be used interchangeably in research and for practical management purposes. The hypotheses are as follows: H1: The three proxies produce similar results with the two fire proxies most likely having a stronger relationship. H2: The data sets can be used interchangeably both for science purposes and in practical management of PAs. The effectiveness of Malagasy protected areas established in or before 2005 (N=42) was examined from 2005 to 2017. Three binary response variables were compared: forest loss, fire incidence, and burned area. In addition, a continuous forest loss variable was examined. Forested areas and the full landscape were studied separately i.e. estimates were produced for both forested areas only and full landscape (forested areas + other areas). 1-kilometre parcels in a uniform grid were sampled using nearest neighbour Mahalanobis distance matching, controlling for the factors affecting conversion pressures with appropriate covariates: altitude, slope, distance to cities, distance to roads, distance to waterways, and rainfall for forested areas and full landscape, and in addition, distance to forest edge for forested areas. Relative effect, pooled relative effect, and network relative effect were calculated for the binary variables, mean effect for the continuous variable. The effects were calculated on country level, biome level (tropical and subtropical moist broadleaved forests, tropical and subtropical dry broadleaved forests, and deserts and xeric shrublands), and individual PA level. Protected areas appeared to be at least moderately effective, and all variables produced parallel, consistent results on the country and biome level, especially when using pooled relative effect. On average, PAs in tropical and subtropical moist broadleaf forests were most effective in avoiding land-use pressures, the ones in tropical and subtropical dry broadleaf forests slightly less, and the ones in deserts and xeric shrublands most ineffective. There was substantial variation between and inside individual PAs, and in approximately half of the PAs all variables indicate that the given area is significantly effective (α = 0,05). In a little over half of the PAs the effects were mixed, and in forested areas, no PA was indicated to be ineffective by all variables. In full landscape, this was the case for one PA. There were small differences between forested areas and the full landscape in all levels, but they were statistically significant only in a few cases. This study thus suggests that the data sets could be used interchangeably, at least on country and biome level, when conducting matching to assess PA effectiveness in a tropical setting. They could be utilised on individual PA level, too, with certain precautions and understanding of the nature and behaviour of the data. They are well suited for research; however, in practical management forest loss and fire incidence might be more feasible than burned area, due to its certain characteristics (it for example demands quite a lot of processing depending on the use purpose) and accessibility issues.