Detection of signals linked to climate change, land-cover change and climate oscillators in Tropical Montane Cloud Forests

Tropical Montane Cloud Forests (TMCFs) form biodiverse communities that are characterized by frequent occurrence of low-level clouds from which they capture a substantial proportion of their precipitation - here referred to as occult precipitation. TMCFs provide important ecosystem services, in particular the supply of water to their wider surroundings. Throughout the tropics (here 23.5 degrees S to 23.5 degrees N), they are under pressure from deforestation and poor land management which leads to loss of both forest area and species diversity, and reduces their capture of occult precipitation. Climate change may also reduce occult precipitation in TMCFs since the cloud base may lift in response to higher temperatures - the 'lifting cloud-base hypothesis'. These threats to TMCFs are well understood, but their quantitative assessment is hampered by 1) uncertainty in the location and spatial extent of TMCFs and 2) limited availability of representative meteorological data. We use a Random Forest Classifier - informed by topographic data, MODIS vegetation data, TRMM precipitation data and ERA5-Land and MERRA-2 reanalysis products - to estimate the spatial distribution and extent of TMCFs (2.1 x 10(6) km(2) +/- 0.5 x 10(6) km(2)). We analyze temporal changes in climate, tree-cover and greenness of TMCFs over the past two to four decades to detect 1) multi-decadal trends, and 2) associations with the El Nino Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD). Evidence for the 'lifting cloud-base hypothesis' in reanalysis products was inconsistent across the tropics; a lifting of the cloud base during the past four decades occurred for about 20% of TMCFs, predominantly in the Americas and a few locations in Africa, while in Asia a downward movement of the cloud base was found. However, these results in part depend on the bias correction applied to the reanalyses. Changes in TMCF tree cover and greenness varied by continent; in Africa in 50% of TMCFs tree cover declined, whereas TMCFs in the Americas and in Asia exhibited a net increase in tree cover, despite a reduction in tree cover in similar to 20% of these TMCFs. An important limitation of the tree-cover data is that they do not distinguish between natural tree cover and agro-forestry. ENSO signals were more strongly present in precipitation in American and Asian TMCFs, whereas IOD signals were stronger in TMCF temperature and dewpoint temperature across the tropics. ENSO and IOD signals were approximately equally important for precipitation in African TMCFs and in cloud-base height across the tropics. An arbitrary warming of 1 degrees C and a 100 m lifting of the cloud base, in accordance with the 'lifting cloud hypothesis', imposed on the Random Forest classifier showed a decline in the extent of TMCFs in the Americas and Africa, but an increase in Asia - mostly at the expense of evergreen broadleaf forests. The greater vulnerability of TMCFs in Africa may be linked to their more isolated and scattered distribution across the continent and drier conditions compared to a more continuous distribution and wetter conditions in the Americas and Asia.