A New Synthetic Proxy Approach of Improved Empirical Relationship Between Proxy and Predictand in Spatial Downscaling

Presently, the remote sensing precipitation (P), evapotranspiration (E) and temperature (T) datasets are coarser (0.1° - 1.25°), which needs accuracy and spatial resolution enhancement in conjugation of auxiliary spatial variables (proxies) before being used at local or regional scales. The existing high-resolution geospatial and environmental proxies possess relatively weak empirical relationships with P, E, and T for accurate reconstruction at high resolution. This research proposed a new concept of synthetic proxies in the deep neural network (DNN) model, the efficiency distribution-based framework (EDBF) algorithm, and the Random Forest (RF) model to estimate the coarse resolution Global Precipitation Mission (GPM) (0.1° ×0.1°), NOAH Evapotranspiration (NOAH-E) (0.25° × 0.25°), and NOAH Temperature (NOAH-T) (0.25° × 0.25°) datasets. Three weight vectors (DNN-R2, DNN-RMSE, DNN-Avg, EDBF, and RF) were generated based on the relationship between multitemporal GPM, NOAH-E, NOAH-T, and DEM values, respectively. Each weight factor was transformed into an independent proxy for estimating those retrievals. From the results, all synthetic proxies outperformed latitude-based natural proxy to efficiently estimate multitemporal GPM, NAOH-E, and NAOH-T retrievals. Specifically, the DNN-RMSE proxy was the most efficient, followed by the EDBF proxy. This work showed that it is feasible to employ the DNN- and EDBF-based synthetic proxies for estimating coarse resolution precipitation, evapotranspiration, and temperature retrievals with greater accuracy, which can onward be employed in the downscaling of these datasets from coarse to fine resolution.