Remote sensing and field data show complementary functions when predicting forage productivity in heterogeneous native forests

Abstract

Native forests around the world are widelyused for livestock grazing as they offer differentsources of forage. Nevertheless, in heterogeneous forested landscapes, forage productivity drivers are stillunclear to make precise predictions of field receptivity. Our aim is to relate landscape variables with forage productivity in forested landscapes using satelliteand ground-based data. To accomplish this, we harvested 36 enclosures in two Patagonian valleys sampled over three years. The location of the enclosuresencompassed a gradient of altitude and mean annualrainfall, across three vegetation types commonlyused for cattle raising. Using a total of 108 biomasssamples, we estimated five generalized linear models to predict forage productivity using remote sensing and ground (field) data as predictors. The mostimportant variables for predicting forage productivitywere five of remote sensing type (the integrated Normalized Difference Vegetation Index, mean annualprecipitation, vegetation type, slope, slope orientation, altitude) and two of field type (canopy opennessand herbaceous layer coverage).The highest goodnessof fit was obtained when all variables were included(D2=0.71). When ground-based information wascombined with remote sensing data, the goodness offit was higher (D2=0.65) compared with models thatonly used remote data as predictors (D2=0.49). Models obtained based on remote data are a useful toolconsidering that field information may not alwaysbe available. High forage productivity levels can be obtained in high forests or scrubs with varying valuesof canopy openness, without removing the forest. Themodels generated in this work are key for livestockstocking rates adjustment in NW Patagonia forests,and may be also re-estimated with new data in otherregions used for cattle raising worldwide, contributing to the sustainable use of native forests.