4. DISCUSSION
Our results illustrate how physical water vapor tracers, such as stable isotopes in precipitation, allow tracking water sources in atmospheric circulation as they connect evaporation and precipitation, and their concentration indicates the meteorological conditions and the proximity to the main atmospheric moisture sources that originate precipitation, as expected. Although the use of stable isotopes in precipitation does not provide quantitative information about the amount of atmospheric moisture from the sources, it complements the information generated by models, such as in our FLEXPART experiment.  More generally, this information has the potential to be integrated into atmospheric models to diagnose the percentage of moisture originated from the different sources (Wang et al., 2004; Fuka et al., 2014; Arias et al., 2015; Hoyos et al., 2018; Molina et al., 2019; Liu et al., 2020).  Further, the combined use of physical tracers and models are particularly useful in regions with complex topographical and meteorological setups where model results can be highly uncertain.
The consideration of the optimal transport day (when the moisture transference is maximum), instead of the canonical 10-day mean lifetime of water vapor in the atmosphere, allowed us to improve the quantification of contributions from atmospheric moisture sources, highlighting the predominance of terrestrial sources all through the year. The natural scales of distance and time are closely linked to the regional structure of moisture source composition. Therefore, assumptions about uniformity or similarity to a global mean need to be revisited for each region, after consideration of the particular climatic and meteorological features. Conversely, the proper transport scales are footprints for each particular target area and include regional features like the mechanisms involved in the atmospheric moisture transport, orographic features, and distance between the target and source regions. The lack of knowledge of these scales could lead to over (or under) estimations of the degree of contribution from different sources and reduce the forecasting power of teleconnections.
Physical water vapor tracers help identify signals of the origin of incoming atmospheric moisture when the source is not influenced by mixing effects associated with different oceanic and terrestrial sources present in the majority of the seasonal analysis. Our results allowed us to identify oceanic sources connected with the seasonality migration of ITCZ, as well as terrestrial sources such as Amazon and Orinoco basins, that agree with meteorological criteria. Both of these observations are evident in the reconstruction of the D-excess, even when the spatial and temporal coverage of isotopic data is very low for Colombia. Although we have reconstructed a baseline for the isotopic structure of precipitation, it is also evident that a better sampling network is necessary to improve the monitoring network of stable isotopes to produce a more detailed analysis of the moisture transport processes, as they have done in more instrumented areas (Friedman, Smith, Gleason, Warden, & Harris, 1992; Kendall & Coplen, 2001; Bowen, Ehleringer, Chesson, Stange, & Cerling, 2007; Bowen, Kennedy, Liu, & Stalker, 2011).
Our results showed a multiannual moisture convergence average from terrestrial sources of 59% for the Andean region and for the Caribbean region of 56% (Figure 6), which is greater than previous estimations by our group (38% in Hoyos et al., 2018). By using a combination of physical tracers and modeling, we confirm previous modeling results (Hoyos et al., 2018) that indicate how, in the study area, the contribution of terrestrial moisture sources to local precipitation is significant (always greater than 44%), such that most ecosystems and water security for society and the economy may depend on the stability of major regional ecosystems such as the Orinoco plains (8% - 28% per month) and the Northern Amazon (17% contribution). More importantly, our results highlight that most terrestrial moisture originates within the same region (NOSA), with contributions larger than 23% per month in some seasons and up to 40% per month in other seasons.
The fact that a significant proportion of rainfall comes from recycling (Fig. 6), highlights how precipitation and, more generally, water availability in the Andean and Caribbean regions of Colombia could potentially be altered by changes in vegetation and land cover, directly affecting transpiration and atmospheric circulation. This is a clear indication that the region is particularly vulnerable to ongoing widespread ecosystem transformation in the region and the surrounding basins. According to Ruiz‑Vásquez et al. (2020), under scenarios that consider deforested areas of approximately 28% and 38% of the Amazon basin, terrestrial sources reduce their annual contributions to northern South America by an incredible average of 40 and 43%. Likewise, Badger and Dirmeyer (2016), confirm that the rise of air masses over northern South America is inhibited with Amazon deforestation, which could also induce inhibition of precipitation over the region. Similarly, the teleconnections with the Orinoco basin reveal that the regional regime of precipitation is highly dependent on a zone in which raising cattle is one of the main economic sources since this is the predominant activity in the area, occupying more than 50% of the productive territory. The expansion of the areas dedicated to this activity is the main source of deforestation (González-González, Villegas, Clerici, & Salazar, 2021). Overall, our results indicate the importance of the hydrological coupling of terrestrial ecosystems in Northern South America. Particularly in Colombia, rainfed agriculture and hydropower generation are an important proportion of the nation’s economy.
Overall, our results highlight an advantage of considering stable isotopes of precipitation over using only numerical modeling, given that many models can either underestimate or overestimate the amount of atmospheric moisture (Hoyos, 2017). Additionally, many atmospheric models have difficulties to represent the topography of the area (Inse, Poulsen, & Ehlers, 2010). Here, this difficulty is faced by analyzing moisture transport phenomena with stable isotopes in precipitation, precisely due to the inverse correlation between the amount of isotopic composition and altitude (Dansgaard, 1964; Rozanski et al., 1993; Mook, 2002) The simultaneous application of both techniques (physical tracer and atmospheric modeling) results in a better interpretation of the transport of atmospheric moisture. The integration of stable isotopes of precipitation and the spatial-temporal modeling can be an accurate tool that reduces the uncertainty associated with the understanding of the climate system and provides information for foreseeing possible changes in hydroclimatic patterns provenience from sources region (Risi, Bony, Vimeux, & Jouzel, 2010; Sánchez-Murillo et al., 2013; Hu & Dominguez, 2015).
In summary, most studies have focused on atmospheric moisture from oceanic sources (Rueda & Poveda, 2006; Sakamoto et al., 2011; Arias et al., 2015; Yepes, Poveda, Mejía, Moreno, & Rueda, 2019) and although a few show the importance of terrestrial sources (Kumar et al., 2016; Hoyos, 2017; Hoyos et al 2018; Molina et al., 2019; Ruiz-Vásquez, Arias, Martínez, & Espinoza, 2020). Our results illustrate the influence of moisture recycling in two important areas of Colombia, with the largest population in the country, and whose economy and ecosystems vitally depend on water availability for ensuring moisture security. Regional atmospheric moisture composition largely controls ecosystem structure, increasing the vulnerability to climate and environmental change in the study area. Further, these natural systems are directly threatened by human activities such as deforestation and intensive agriculture, altering the exchange of water in the land-atmospheric interactions, the energy balance, evapotranspiration, and therefore, the atmospheric moisture content and transport.