Number of Fires Trend Dashboard - Country level

This layer shows the trend of fires occurrence at country level (UN_0 UN Cartographic Section level 0) from 2003 to present based on MODIS Collection 6 Active Fire Product MCD14ML. The dashboard associated to the layer provides relevant information over 1) the density of fires by Km2 per each administrative unit, 2) the evolution of the monthly cumulated fires per each year, 3) the distribution of fires by land cover type (based on MODIS landcover data per each year) and 3) the number of fires occurring in critical areas such as forests, protected areas (WDPA) and forests in protected areas.

The administrative units are identified using UN Cartographic Section's UN_0 polygon layer for countries. Point data events from NASA were associated to the administrative units and the temporally coherent land cover through a spatial jointure in GRASS GIS. Before the trend is calculated, data were pre-processed in order to obtain the fire density (per km2) of the yearly cumulated fires in each administrative unit. Only the fires attributed to natural events were used for the computation of the trend (only events with "type==0" in NASA's dataset were used if near-real time data had to be used). The computation of the trend is based on the Theil–Sen estimator - a robust method for linear regressions that is insensitive to outliers. The resulting positive and negative values of the slope were then normalized separately to a set of values ranging from -3 (very decreasing trend) to +3 (very increasing trend) using a standard deviation based distribution of the data across different administrative units. Thresholds for normalization were set at 0.1STD, 0.5STD and 1*STD where STD is the STD calculated on the positive subset of the slope values for positive values and the STD calculated on the negative subset of the slope values for negative values. Data showing a slope estimate below the first positive and negative threshold are given a value of 0 indicating a stable trend. The same value (0) is also given to those whose trend is not significant using a threshold of 90% (pvalue of Kendall rank correlation coefficient = 0.1).

The trend was calculated in R using, especially, the function zyp.sen of the package zyp (for magnitude and intercept of the Sens slope) and MannKendall of the package Kendall (to determine the significative value of the trend through Kendall rank correlation coefficient).