Adapting transition matrix
The Trend.Earth platform by default uses expert matrices to diagnose trends in land cover change (degradation, stable, improvement) and productivity dynamics (improving, stable, stable but stressed, early decline, declining). The experience of using default matrices for pilot territories (administrative units or individual farms) revealed that the greatest challenge was interpretating transitions between the categories of tree-covered, grasslands, croplands and artificial lands (Ptichnikov and Martynyuk 2020), (Kust et al. 2019). Tree-covered lands, as a result of felling in managed forests, often turn into grasslands and then gradually recover. The recovery successions can continue for dozens and hundreds of years and can be further complicated by stages of wetlands and grasslands. For example, as a result of abandonment (Kust, Andreeva, and Cowie 2017) of agricultural land in Russia in late 80s - 90s (there are up to 50 million hectares of abandoned farmland) (National Report 2019), significant areas remain covered with forest vegetation, while in southern regions they turned into unproductive pastures. For these previously productive lands such transformations are negative from an economic viewpoint. In contrary, their return to farmland started in 00-s is characterized by a formal decrease in NDVI, but considered as a positive process from an economic standpoint. Likewise, many of the depleted peatlands in drained wetlands are hot spots for wildfire risks despite possible turning to forests or meadows. Therefore, their artificial re-wetting should be considered a positive change. Finally, many residential areas, especially in arid regions, are largely covered with green vegetation and have a higher productive potential in contrast to the adjacent natural landscapes. Hence, the reconsidering of evaluation matrices is needed, but it should also be respective different regions. For example, for individual farm in the Samara region it was demonstrated that a locally adapted matrix can change the assessment results from negative to positive (RSF, 2020). For pilot sites in the Kaliningrad region the opposite situation was observed: with the positive assessment by default, the matrix adjustment and use of additional indicators calculated the PDL to be that of 86.6%! (Makarov et al. 2021). For the test farm in the Penza region, an increase of the area of degraded lands was also noted after changing the assessment matrix (Makarov et al. 2021a). In addition to tuning transitions between major land types, in many cases it is important to add specific categories (land subtypes). For example, for bushing areas in the Samara and Kursk regions (forest-steppe subzone) it was important to distinguish young (up to 10-15 years old) and mature (over 15-20 years old) forests (IGRAS 2021). For sites in the Volgograd region (dry steppe subzone) the division into irrigated and non-irrigated arable land was important. For the semi-desert Trans-Volga region, it was essential to subdivide pastures by the degree of degradation (Slavko, Kust G. and Andreeva 2022). Ptichnikov et al. 2019) proposed establishing categories for lands with different types of forest regeneration dynamics in the Komi forests (for example, overgrown burned areas and clearings).
Therefore, the reconstruction of transition matrix is an important element to identify critical land cover and land productivity trends. It should be carried out selectively for various locations at local and provincial levels respecting local conditions and adding land-specific sub-categories.