Analysis and integration of regional scale temperature datasets into a seasonal crop monitoring system.

Abstract

Populations in excess of 20 million people in southern Africa annually face food insecurity. This number increases appreciably when detrimental seasonal climate conditions lead to widespread reductions in crop harvests. This situation has led to the development of regionalscale crop monitoring systems that incorporate crop-specific water balance (CSWB) models for early detection and warning of impending weather-related crop production shortfalls. Early warning of anticipated reductions in crop harvests facilitates early action in responding to potential crises. One such system, used by the Famine Early Warning Systems Network (FEWS NET) in southern Africa for crop monitoring, calculates the water requirements satisfaction index (WRSI) using a CSWB model. Operationally, CSWB models for calculating WRSI have used a static length of growing period (LGP) to bracket the period over which rainfall and evapotranspiration variations can affect crop yields. In the long term, concerns have been raised by some studies on the impact of rising air temperatures on crop production. There is therefore a need to incorporate the impacts of air temperature on crops directly into food security monitoring systems, in order to improve the accuracy of these monitoring systems in identifying and locating weather-related crop production shortfalls. This study sought to assess the potential improvements that can be introduced to the crop monitoring system in general and the WRSI in particular, by incorporating air temperature data into the CSWB model. To address this objective, daily maximum and minimum air temperature grids derived from a general circulation model reanalysis were used to generate thermal time estimates, expressed as growing degree day (GDD) grids for a maize crop. The GDDs were used to estimate the LGP of maize for each pixel of each summer season (which typically runs between around October and March) from 1982/1983 to 2016/2017 in southern Africa. The variable, temperature-driven LGP estimates compared favourably with LGP values obtained from literature for a few sample locations. The variable LGP was used to calculate the WRSI for 35 seasons, and the resultant WRSI showed improved correlation with historical yield estimates compared to the static-LGP WRSI, particularly after the farming practice of planting on multiple dates was taken into consideration. Various expressions of WRSI were considered in the analysis, including WRSI calculated assuming planting at the onset of rains, WRSI aggregated from varying number of separate planting dates, including three and six planting dates as test examples, and WRSI calculated using a modified soil water holding capacity to better capture local soil management practices. Historical maize yields for sub-national administrative units from seven southern African countries were correlated with the various WRSI expressions. Gridded GDD data that reflect the accumulated severity of extreme warm temperatures experienced during the crop growth period, referred to as extreme growing degree days, or eGDD, were also noted to have significant correlations with historical maize yields in several southern African countries. A number of variants of the eGDDs were tested, including eGDDs accumulated throughout the crop’s growth period, eGDDs that only occurred simultaneously with periods of crop water deficit, eGDDs that occurred during the crop flowering stage, and eGDDs scaled by the severity of crop water deficit. In several areas, the various eGDD expressions indicated higher correlations with yield than any of the WRSI variants indicated. The eGDD parameter showed strong correlations with WRSI, suggesting that the accumulated high temperatures were a reflection of the influence of low rainfall and low soil moisture during episodes of high temperature. More work is required to calibrate and refine the temperature-based monitoring parameters that were developed in this study, at local, sub-national scales. In particular, assumptions of the linearity of maize yield response for the various parameters should be tested. Potential improvements of the combined eGDD-water deficit parameter through the incorporation of prediction coefficients and constants should also be tested. A secondary aim of the study was to explore how readily available temperature-related datasets can be utilized to derive air-temperature metrics. To this end, satellite-derived thermal infrared (TIR) brightness temperature data were analysed, and a method was developed for identifying cloud cover, while simultaneously estimating cloud-free diurnal brightness temperature curves, using a single TIR satellite channel. The diurnal brightness temperature curves were developed using a sinusoidal and exponential model for daytime and nighttime respectively, utilizing modifications that enabled the curves to be estimated from two known temperatures at any two given times with cloud-free brightness temperature scenes. Comparison of the cloud mask developed in this study with an existing operational cloud mask based on a methodology developed by the EUMETSAT Satellite Applications Facility for Nowcasting gave an accuracy of 85.4%, when the operational method was considered as truth in a confusion matrix analysis. Situations were identified in which the different cloud detection methods showed superior performance, and could therefore complement each other. A statistical method was also developed for calibrating the cloud-free brightness temperatures to station-observed 2-m air temperatures using relationships between the means and diurnal temperature ranges of the two datasets. This enabled the identification of periods of occurrence of extreme warm air temperatures with a coefficient of determination of 0.91, and demonstrated the potential for the usage of TIR data for generating estimates of useful air temperature metrics. The efficiency of the algorithms that were used for simultaneous cloud masking and generation of cloud-free brightness temperature should be improved, in order to enable the methodology to be scaled up to a regional or global gridded level of analysis. Further work for improving operational gridded air temperature datasets by combining station-observed temperature data, modelled data from global circulation models, satellite-derived modelled cloud-free brightness temperature data and cloud masks is recommended.