Thus, innovative ways to estimate catchment rainfall and to improve the estimation of catchment design rainfall are required. Despite the varied performance, the result of the study shows that CHIRPS rainfall product can be used to estimate catchment rainfall for hydrological modeling and flood frequency analysis.
INTRODUCTION
The main purpose of this study is to use remotely sensed rainfall to improve catchment rainfall estimation for use in hydrological modeling and evaluation of different methods of catchment rainfall estimation and catchment design rainfall. This is to (a) address the challenge of few stations with long records of rainfall data and (b) address the gap that exists in deriving catchment design rainfall for flood studies in less gauged catchments.
LITERATURE REVIEW
- Rainfall Observation and Monitoring Network
- Daily raingauge observations
- Raingauge network status in South Africa
- Estimation of Catchment Design Rainfalls
- Using observed point rainfall data
- Using remote sensing
- Design rainfall estimation
- Verification and Bias Correction of Remotely Sensed Data
- Chapter Summary
Therefore, bias correction is recommended before applying remotely sensed rainfall data (Habib et al., 2014a; Maswanganye reported the erratic performance of RS products. Research by Bhatti et al. 2016) indicates that satellite-based rainfall estimates are not always reliable. .
STUDY METHODOLOGY
- Catchments Selection
- Rainfall Station Selection
- Remote Sensing Product Selection
- Extraction and Processing of Remote Sensing Rainfall Data
- Using ArcMap
- Using Python
- Using Google Earth Engine to extract point pixel values
- Using Google Earth Engine to extract mean catchment rainfall
- Performance of RS Data
- Bias Correction
- Estimation of Daily Catchment Rainfall
- Using observed rainfall: Weighted stations
- Using remotely sensed rainfall
- Estimation of Catchment 1-Day Design Rainfall
Details of the selected stations are summarized in table 3.1 for catchment S60A and table 3.2 for catchment U20F. However, there is uncertainty with the daily time period used in the development of the CHIRPS data set, that is, the performance of the selected RS product will be evaluated and if necessary bias corrected for the selected local catchments.
The equations used to estimate the values of the selected statistics are presented below: i) Mean bias error (MBE) is mainly used to estimate the mean bias in the model and captures the mean bias in the forecast. Using the spatial distribution of the selected stations, Thiessen polygons were constructed, and the area weights of each station were estimated. An example of the calculations section is shown in the spreadsheet found in Appendix E Table E.2.
The results of the estimation of the median amount of precipitation are presented in Table E.3 of Appendix E for the S60A catchment and Table E.4 for the U20F catchment. The Thiessen weights of the selected stations are equal to the weights of the corresponding pixels. To evaluate the design of catchment rainfall using estimated catchment rainfall estimated using different methods, R code was used to fit a generalized extreme value (GEV) distribution to the annual maximum set (AMS) of catchment rainfall data using L -moments. (Smithers et al., 2018).
APPLICATION TO CATCHMENT S60A: PILOT STUDY
- Assessment of the performance and bias correction
- Sensitivity to Period Length
- Estimation of Catchment Rainfall
- Validation of method of catchment rainfall correction
- Assessment on a daily scale
- Assessment on a monthly scale
- Estimation of Catchment 1-Day Design Rainfall
Given the presented approaches being compared, it is clear that Weighted pixels have estimates that are much closer to the accumulated daily rainfall of the Weighted station followed by the Driver pixel (pptcor) approach. This could be the result of a similar approach to estimating catchment rainfall used in both the Weighted Pixel Approach and the Weighted Station Approach and the point rainfalls were obtained from the same geographic areas. Furthermore, the Driver pixel (pptcor) approach performs well because it is based on the selection of a pixel that best represents the watershed and therefore has a high contribution to the rainfall derived from the weighted pixel.
This can be seen in the summary of the MAE statistics in Table 4.2, where the weighted pixel had a low MAE value and the corrected GEE and driver pixel (pptcor) had the values of MAE 8.57 and 8.56, respectively. In terms of distribution performance, the Weighted Pixel approach shows a similar distribution to the Weighted Station approach, while the distributions of the Corrected GEE and the Driver pixel (pptcor) approaches are less similar. Based on the graph, the approximations performed better from the weighted pixels followed by the corrected GEE with similar performance.
Closer analysis of the design values shows varying performance of all approaches relative to the weighted station approach. However, the MAE estimates in Table 4.4 show that the Weighted pixel approach has the lowest estimated error, while the Driver pixel (pptcor) has the highest estimated error. So while all approaches performed reasonably well, the Weighted Pixel approach performed best.
APPLICATION TO CATCHMENT U20F
Assessment of the performance and bias correction
The results of the differences in MBE between RAW and CORRECTED are shown in Table 5.1. Further assessment of the effect of spatial density on catchment rainfall estimation was undertaken using the regression graphs shown in Figure 5.11. Therefore, without an outlier, the results prove that the weighted pixel approach is the best estimate of rainfall in the catchment design.
In the case of the bias correction results, the EQM approach was used in both catchments in this study. In terms of temporal scale, the performance of the bias correction method improved from daily scale to annual scale. The results also showed a lack of enhanced bias in the RS values at two of the stations in Catchment U20F.
The results showed that the performance of bias correction using the EQM technique is not significantly affected by the change in period length. Estimated 1-day catchment rainfall design results were compared for the same approaches that were used to estimate catchment rainfall for the S60A catchment and the U20F catchment, with the addition of the steering pixel (ARF). The GEE-corrected approach also performed well with a low MAE value compared to the station-weighted approach.
Catchment Rainfall Estimates
- Assessment on a daily scale
- Assessment on a monthly scale
- Sensitivity to the spatial density and influence of Thiessen weights
- Assessment using daily accumulation
- Assessment using regression statistics
Estimation of Catchment 1-Day Design Rainfall
The catchment design rainfall results estimated for Catchment S60A suggested better performance in estimating catchment design rainfall by the Weighted pixel rainfall approach, even though the other approaches (Driver pixel(pptcor) and Corrected GEE) still performed well, and the assumption was that the results will be the same in Catchment U20F. The main differences in catchments S60A and U20F are therefore the different climate regions and the distribution of stations over the catchment, which has an impact on Thiessen weights. The results are presented in Figure 5.12 which shows the design values for each return period.
Based on the plot, the approaches performed better compared to the weighted stations approach by the weighted pixel approach, the corrected GEE, the steering pixel (ARF) and the steering pixel (pptcor). The guiding pixel (pptcor) and guiding pixel (ARF) approaches overestimate the design values for most return periods while the corrected GEE is slightly underestimating the design values for all return periods except the 10-year return period. However, the MAE estimates in Table 5.4 show that the corrected GEE has the lowest estimated error followed by the weighted pixel approach while the driver pixel (pptcor) has the highest estimated error.
The switch between the GEE-corrected and pixel-weighted approach in terms of which one has the best rating is influenced by the observation of an observed design extreme value of 240 mm in the 20-year return period, which resulted from a value of external that was observed in the observed precipitation. Dataset in March 2000. The MAE between weighted stations and the outlier-free weighted pixel approach is 5.648.
DISCUSSIONS, CONCLUSIONS, AND RECOMMENDATIONS
In this study, catchment rainfall was estimated using four approaches: weighted stations, weighted pixels, pixel leader (pptcor) and the corrected GEE approach. A sensitivity analysis was done in the U20F catchment on the spatial density of the station to estimate the catchment's rainfall using the weighted station and weighted pixel approach. Estimates of catchment precipitation using three stations were compared with the estimate of catchment precipitation using the original selected stations (10).
This may be because the 10 stations were not evenly distributed and the 3 selected stations used had a higher weighted contribution to the catchment's total rainfall when using the 10 stations. Overall, the results demonstrate the ability of local bias-corrected RS CHIRPS data to provide accurate catchment rainfall estimates, because the results between catchment rainfall estimated using 3 and 10 stations from the weighted station approach are similar with results obtained using the weighted station approach. Catchment rainfall from a single station with multiple ARFs is commonly used in practice as the traditional method to estimate catchment design rainfall from a single point, however, this approach (using the pixel point value) performed well only in the S60A catchment compared to the U20F catchment.
Although weighted stations were assumed to be the best estimate of catchment rainfall in this study, there are limitations to this approach and it will work best when there is a good spatial density of gauges. This project aimed to use remotely sensed rainfall to improve catchment rainfall estimation for use in hydrological modeling and the evaluation of different catchment rainfall estimation methods and catchment design rainfall. The bias-corrected CHIRPS RS product was used to estimate the rainfall of the remotely sensed catchments and, although the performance of the RS products is location dependent, CHIRPS performed well for the selected study catchments.
Evaluating the Accuracy of Multisatellite GPM Precipitation Products in Hydrological Applications in Alpine and Gorge Regions with a Sparse Network of Rain Gauges. Effect of bias correction of satellite rainfall estimates on runoff simulations at the source of the Upper Blue Nile. Advances in the application of the optimal interpolation method to the renovation of the Zambian rain gauge network.
Assessment of hydrological performance of multiple satellite rainfall products in the upper Blue Nile Basin, Ethiopia. A review of the use of Earth remote sensing in water resources management in South Africa. Comparison of remotely sensed rainfall estimates with observed data from rain gauges in the Western Cape, South Africa.
CalcPPTCor: A tool to assist in rain station selection and adjustment of rainfall data. Using satellite-derived rainfall estimates to extend water resources simulation modeling in South Africa. Analysis of regional flood frequency and spatial patterns in the Pearl River Delta region using the L-moment approach.
APPENDIX A: PYTHON CODE TO EXTRACT CHIRPS RAINFALL DATA
APPENDIX B: PYTHON SCRIPT TO EXTRACT CHIRPS RAINFALL DATA
APPENDIX C: CODE TO EXTRACT RESAMPLED CHIRPS DAILY
Import image collections, filter by date and ROI, apply cloud mask and clip to ROI. Export the average rainfall for each polygon as a .csv file Export.table.toDrive({ . collection: finalRainfall, . description: 'CHIRPS_Rainfall'+startdate+'TO'+enddate, folder: 'Genus_Exchange_GEE_Data', . fileFormat: 'CSV '.
