Hurricanes Maria and Irma (IrMaria) hit Puerto Rico in September 2017, causing widespread damage to forests that can be expected to have significant consequences for water resources. Here we used estimated watershed-level vegetation damage and biomass metrics to explain changes to streamflow (ft3/s) from pre-hurricane levels. Watersheds were delineated for 48 USGS daily streamflow monitoring stations with more than 50 values after Maria hit, and zonal statistics for damage and biomass were estimated. To estimate damage, spectral mixture analysis was applied to extract per-pixel proportions of non-photosynthetic vegetation (NPV) from before IrMaria (September 1 – November 30, 2016) and after (September 20 – November 30, 2017). Changes in the proportion of non-photosynthetic vegetation (ΔNPV) between composites were used to represent hurricane damage. LiDAR-derived canopy height was tested against field-estimates to be used as a proxy for biomass. Post-hurricane streamflow values were calculated as the deviation from the site-level long term average (2005-2017) and linear mixed effect models were used to examine fixed effect predictors including precipitation, mean watershed ΔNPV, and watershed-level total biomass. Site was included as a random effect.
Results/Conclusions
In the subsequent days after IrMaria, streams experienced drastic increases in streamflow, often orders of magnitude higher than average. This was followed by a steep decline in values that stabilized at variable levels relative to the long-term average. All watersheds showed some level of damage (i.e. increase in NPV) after the hurricane, and watershed-level ΔNPV ranged from 3%-59%. After finding a strong positive linear relationship (R2=0.78) between field-based biomass and canopy height, the sum of all canopy height pixel values within a watershed was used to represent total watershed biomass. Using watershed-level mean ΔNPV, sum of canopy height, and daily precipitation as predictors, the model accounted for 15% of the variation in deviation from the site-specific long-term average streamflow seen after the hurricane. Among predictors, precipitation had the greatest effect (0.059 ± 0.002). Increased average watershed damage also positively impacted deviation from long-term average streamflow (0.018 ± 0.006), while greater amounts of biomass within a watershed decreased streamflow (-0.021 ± 0.005). While our results indicate that precipitation remains the largest predictor of streamflow, watersheds with high amounts of biomass generally experience lower streamflow values. High levels of vegetative damage significantly increases streamflow relative to pre-hurricane levels.