Satellite-based rainfall products assessment for model predictions of rice yield

Rainfall is one of the drivers of rice crop growth where the amount and the duration of rainfall can affect the rice crop yield. Accurate estimation of rainfall event is important in order to generate accurate rice yield predictions. In this study, rainfall estimates from satellite are evaluated prior to use for rice crop modelling. The satellite based rainfall (SBR) products assessed are the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG), Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS) and Climate Precipitation centre MORPHing (CMORPH) from 2015-2017. A comparative analysis is carried out between the SBR estimates and rainfall estimates from 898 rain gauge stations in Malaysia. The performances are evaluated in terms of detection and volume metrics. Detection metrics include Probability of Detection (POD), False Alarm Ratio (FAR) and Threat Score (TS) while the volumetric metrics include Root Mean Squared Error (RMSE), Correlation of Coefficient (CC), Percent Bias (PBIAS) and Nash Sutcliffe Error (NSE). Satellite-rain gauge data adjustment is carried out to improve the accuracy of rainfall estimation. The adjustment methods used are Mean Bias Correction (MBC), Residual Inverse Distance Weighting (RIDW), Double-kernel Smoothing (DS) and Kriging with External Drifts (KED). After adjustment, both pre-adjusted products and post-adjusted products are used to drive a rice crop model called ORYZA (v3). The resulting estimations of rice physiological parameters such as Leaf Area Index (LAI), total above ground biomass (WAGT), dry weight of stems (WST), dry weight of green leaves (WLVG), dry weight of dead leaves (WLVD) and dry weight of storage organs (WSO) are compared against estimates of the baseline model which is driven by the weather station data installed at IADA Ketara (Northern Terengganu Integrated Development Project) as well as field observations. The results show that pre-adjustment, CHIRPS has the lowest average error (RMSEave = 15.95 mm/d), while IMERG has the highest correlation (CCave =0.36) and best percentage bias (PBIASave = 11.86%) values. Average NSE values were below zero for all three pre-adjusted products. In addition, IMERG shows the best detection skills (PODave = 0.98, TSave= 0.64) while CMORPH shows the lowest false alarm rate (FARave = 0.32). For the adjustment methods, KED adjusted rainfall products showed the best improvement followed by DS, RIDW and MBC. The CMORPHKED has the lowest average error (RMSEave = 11.66 mm d-1), highest correlation (CCave = 0.66) and highest NSE values (NSEave = 0.43) while IMERGKED has the best percentage bias (PBIASave = -1.59%). In the ORYZA (v3) application, all models driven by satellite products show a small change from the baseline model. CMORPHMBC have similar RMSE (LAI = 0.34, WAGT = 3429.05 kg ha-1, WST = 1144.59 kg ha-1, WLVG = 904.08 kg ha-1, WLVD = 537.95 kg ha-1 and WSO = 1254.02 kg ha-1) and NSE (LAI = 0.88, WAGT = 0.38, WST = 0.61, WLVG=-3.65, WLVD=0.22 and WSO = 0.70) results as the baseline model. Based on the analysis of ORYZA (v3) simulations with and without irrigation, as well as outputs of surface runoff, the model is found to be sensitive to rainwater input but not to the differences in the amounts between SBR products as the water provided to the crop is in surplus. Based on the results obtained, IMERG have the best performance prior to adjustment. The performance of all SBR products improved after being adjusted. Kriging with external drifts shows the best improvement compared to other adjustment methods. For ORYZA, the model could not discriminate performance of pre-adjustment and post-adjustment SBR because water supply was always in excess due to the location of case study where it tends to receive high amount of rainfall.

Text FK 2020 76 ir.pdf Download (1MB)

Actions (login required)

View Item