- Email: [email protected]
Peer review report 2 on “Assimilating a synthetic Kalman filter leaf area index series into the WOFOST model to improve regional winter wheat yield estimation”
Agricultural and Forest Meteorology 201S (2015) 664
Contents lists available at ScienceDirect
Agricultural and Forest Meteorology journal homepage: www.elsevier.com/locate/agrformet
Peer review report
Peer review report 2 on “Assimilating a synthetic Kalman filter leaf area index series into the WOFOST model to improve regional winter wheat yield estimation”
1. Original Submission 1.1. Recommendation Minor Revision 2. Comments to Author I have reviewed the manuscript entitled “assimilating a synthetic Kalman filter leaf area index time series into the WOFOST model to improve regional winter-wheat yield estimation. The manuscript presents an interesting approach to the well-known mismatch between the spatial resolution of satellite observations and the landscape scale of many agricultural landscapes. A timeseries of 1-km resolution MODIS observation is combined with a few LandSAT TM images to obtain a synthetic time-series of LAI observation at TM resolution. Despite some deficiencies, the authors demonstrate that the synthetic LAI time series can be used to adjust the WOFOST model using an ensemble Kalman filter and that this improves estimates of LAI and yield at 1 km and at regional scale. In general the paper is well written and can be accepted for publication with minor revisions. The manuscript needs some carefull editing as there are some typos, but in general I only have a few comments that need to be solved: L292, 294: there are some super/subscripts characters that are missing here. L318-322: the explanation of the model used to connect the MODIS data with the TM data is too short. Although you refer to Sedano et al it needs a bit more introduction for readability. L332: I do not understand the sentence “only the result of the measurement update of the previous time step”. If you do not have a measurement, then your model will propagate forward without any analysis step. So I do no see what you mean to say here. L335: How are the forward and backward model combined? L417: I find this equation rather fuzzy. In my understanding ‘E’ is an inflation factor to the model covariance in order to avoid fil-
DOI of original article: http://dx.doi.org/10.1016/j.agrformet.2015.10.013. 0168-1923/$ – see front matter http://dx.doi.org/10.1016/j.agrformet.2015.12.044
ter divergence. But where does the ‘150’ come from? I asssume that this is ‘150 days’ (as k means time) and is meant to gradually increase the inflation factor, so after 150 days of running time the inflation factor is maximal. Anyway this needs some better explanation. L419-437: This description leaves me wondering how exactly your WOFOST/EnKF is implemented? In line 432 you state that the mean of the LAI ensembles is used as input to the WOFOST model. If this the case then your implementation is not a ‘True’ EnKF because an EnKF updates the LAI of each ensemble member with the posterior estimate and then propagates the entire ensemble. In your implementation you seem to propagate a single WOFOST member with the mean LAI of the ensemble. Although this does not invalidate this study, it must be mentioned how exactly it is done. L454: I doubt whether this should be called ‘uncertainty’ maybe ‘variability’ is a better term. L475: why you use ‘spatial’ here, I think it should be ‘temporal’ as you are looking into the temporal dynamics. L555: seen instead of seem L621-622: this sentence is not good, please fix it. L680-696: first of all, this section is out of place here. Second, I doubt if it really applies to your case because the study of Curnell et al dealt with the case that you did not know the true sowing date of the crop. In such cases you end up with shifts in phenology between the model and the satellite observation which can deteriorate your results. However, you seem to have a good estimate of the true sowing date. Figure 8 The readability of these charts must be improved: make them square with the range the same on the x and y axes across all charts. Moreover, add an 1:1 line. That way the biases will be immediately visible. Anonymous Available online 18 December 2015
Contents lists available at ScienceDirect
Agricultural and Forest Meteorology journal homepage: www.elsevier.com/locate/agrformet
Peer review report
Peer review report 2 on “Assimilating a synthetic Kalman filter leaf area index series into the WOFOST model to improve regional winter wheat yield estimation”
1. Original Submission 1.1. Recommendation Minor Revision 2. Comments to Author I have reviewed the manuscript entitled “assimilating a synthetic Kalman filter leaf area index time series into the WOFOST model to improve regional winter-wheat yield estimation. The manuscript presents an interesting approach to the well-known mismatch between the spatial resolution of satellite observations and the landscape scale of many agricultural landscapes. A timeseries of 1-km resolution MODIS observation is combined with a few LandSAT TM images to obtain a synthetic time-series of LAI observation at TM resolution. Despite some deficiencies, the authors demonstrate that the synthetic LAI time series can be used to adjust the WOFOST model using an ensemble Kalman filter and that this improves estimates of LAI and yield at 1 km and at regional scale. In general the paper is well written and can be accepted for publication with minor revisions. The manuscript needs some carefull editing as there are some typos, but in general I only have a few comments that need to be solved: L292, 294: there are some super/subscripts characters that are missing here. L318-322: the explanation of the model used to connect the MODIS data with the TM data is too short. Although you refer to Sedano et al it needs a bit more introduction for readability. L332: I do not understand the sentence “only the result of the measurement update of the previous time step”. If you do not have a measurement, then your model will propagate forward without any analysis step. So I do no see what you mean to say here. L335: How are the forward and backward model combined? L417: I find this equation rather fuzzy. In my understanding ‘E’ is an inflation factor to the model covariance in order to avoid fil-
DOI of original article: http://dx.doi.org/10.1016/j.agrformet.2015.10.013. 0168-1923/$ – see front matter http://dx.doi.org/10.1016/j.agrformet.2015.12.044
ter divergence. But where does the ‘150’ come from? I asssume that this is ‘150 days’ (as k means time) and is meant to gradually increase the inflation factor, so after 150 days of running time the inflation factor is maximal. Anyway this needs some better explanation. L419-437: This description leaves me wondering how exactly your WOFOST/EnKF is implemented? In line 432 you state that the mean of the LAI ensembles is used as input to the WOFOST model. If this the case then your implementation is not a ‘True’ EnKF because an EnKF updates the LAI of each ensemble member with the posterior estimate and then propagates the entire ensemble. In your implementation you seem to propagate a single WOFOST member with the mean LAI of the ensemble. Although this does not invalidate this study, it must be mentioned how exactly it is done. L454: I doubt whether this should be called ‘uncertainty’ maybe ‘variability’ is a better term. L475: why you use ‘spatial’ here, I think it should be ‘temporal’ as you are looking into the temporal dynamics. L555: seen instead of seem L621-622: this sentence is not good, please fix it. L680-696: first of all, this section is out of place here. Second, I doubt if it really applies to your case because the study of Curnell et al dealt with the case that you did not know the true sowing date of the crop. In such cases you end up with shifts in phenology between the model and the satellite observation which can deteriorate your results. However, you seem to have a good estimate of the true sowing date. Figure 8 The readability of these charts must be improved: make them square with the range the same on the x and y axes across all charts. Moreover, add an 1:1 line. That way the biases will be immediately visible. Anonymous Available online 18 December 2015